# Asymmetri > Asymmetri is a hard problems technology company. We engineer AI systems, agents and digital products for global brands, across intelligence, process, experience and scale. ## Key facts - Name: Asymmetri (legal entity: Asymmetri Innovations LLP) - Category: Hard problems technology company; AI engineering and agent development company - Tagline: Global brands. Hard problems. Solved. - Positioning: The technology company for what's next. - Founder and CEO: Nandagopal P (https://www.nandagopalp.com · https://www.linkedin.com/in/nandagopal-p/) - Based in: Kerala, India. Remote-first team working with clients worldwide - Practices: AI Agentic Systems and Data; Application Development; Technology Consulting; Innovation and R&D - Industries: Commerce and Finance; Digital Experience, Media and Growth; Health, Care and Wellbeing; Operations, Industry and Logistics; Physical and Connected Environments - Engagement model: Frame → Design → Engineer → Prove → Operate, from prototype to pilot to production - Model-agnostic: OpenAI, Anthropic Claude, Google Gemini and open models, chosen per use case - Contact: hey@asymmetri.in · WhatsApp https://wa.me/919847206822 · https://www.asymmetri.in/contact-us - Careers: https://www.asymmetri.in/careers - Official profiles: LinkedIn https://www.linkedin.com/company/asymmetri/ · Instagram https://www.instagram.com/asymmetri.in/ · X https://x.com/Asymmetri_ ## Who we are Asymmetri is a hard problems technology company. We solve consequential problems across four areas: - **Intelligence**: Systems that think, decide and act. Models, knowledge, memory and agents engineered into one accountable layer. - **Process**: Operations redesigned around intelligence, so work is planned, coordinated and completed with people in control. - **Experience**: Products people transact with, depend on and trust, on screens and in the physical world. - **Scale**: Infrastructure, governance and unit economics that hold up from pilot to production. ## Founder Asymmetri was founded by Nandagopal P (founder and CEO; https://www.nandagopalp.com; LinkedIn https://www.linkedin.com/in/nandagopal-p/), an engineer, venture builder and TEDx speaker. The name comes from an idea he returns to: existence itself began as a small asymmetry, and what endures usually comes from one small, true insight others stepped past. He pairs that operating view with a reflective one (attention, intention and conscious ambition), summed up in the line he quotes from the Chandogya Upanishad: Tat Tvam Asi, "You are That". ## What we do: four practices 01. **[AI Agentic Systems and Data](https://www.asymmetri.in/what-we-do/ai-agentic-systems-and-data)**: Systems that think, decide and act on your data. Agents, knowledge and data engineered into one accountable operating layer. From retrieval and memory to orchestration, evaluation and governance, we build AI that leaves the chat window and does the work. Capabilities: Agent runtimes, Multi-agent orchestration, RAG and knowledge graphs, Data platforms and pipelines, Evaluation and observability, Guardrails and governance. 02. **[Application Development](https://www.asymmetri.in/what-we-do/application-development)**: Products people transact with, depend on and trust. Web platforms, mobile apps, SaaS and connected devices, designed and engineered by one team. An AI-driven delivery loop takes the product from first commit to production, then keeps it improving. Capabilities: Web platforms and SaaS, iOS, Android and cross-platform, Product and interaction design, APIs and integrations, Cloud, DevOps and scale, QA and test automation. 03. **[Technology Consulting](https://www.asymmetri.in/what-we-do/technology-consulting)**: Frame the hard problem. Architect the answer. Strategy, architecture and technical leadership from people who also build. We find where technology creates real operating leverage, design the system, and stay accountable until it runs. Capabilities: Problem framing and discovery, AI strategy and readiness, System architecture, Technical due diligence, Fractional technical leadership, Delivery recovery. 04. **[Innovation and R&D](https://www.asymmetri.in/what-we-do/innovation-and-rd)**: What is next, engineered into what ships. Applied research in AI and machine learning systems and in mixed reality, plus the products we build and run ourselves. We prototype the frontier, prove it, then transfer it into production. Capabilities: Applied AI and ML research, Mixed reality and spatial UI, Rapid prototyping, Benchmarks and evaluation, New product incubation, Research to production transfer. ## Point of view AI becomes valuable when it leaves the chat window and enters the operation. - From interface to interaction, judgment and approval - From information to contextual, retrievable knowledge - From automation to agents that plan, coordinate and act - From experiment to governed, observable operating capability We do not add AI around the edges. We redesign the system around intelligence. Intelligence is not a feature. It is a new operating layer. ## The Asymmetri AI system: ten layers One connected path from idea to operating leverage. Technologies listed are representative, not a fixed stack. 1. **Design** (AI product and interaction design): We design AI people can understand, direct and trust. Capabilities: Conversational UX, Agent interfaces, Copilots, Generative UI, Human-in-the-loop review, Approval experiences. Representative stack: AI SDK UI, CopilotKit, assistant-ui, AG-UI Protocol, OpenAI ChatKit, Gradio, Streamlit. 2. **Knowledge** (Data and knowledge architecture): We turn fragmented information into retrievable operating context. Capabilities: Databases, Ingestion and retrieval, Embeddings, Vector stores, Knowledge graphs, RAG pipelines. Representative stack: PostgreSQL, Pinecone, Qdrant, Neo4j, Weaviate, Unstructured, Apache Kafka, Elasticsearch. 3. **Intelligence** (Model and inference engineering): We engineer the intelligence layer around quality, speed, control and cost. Capabilities: Model selection, Model routing, Structured outputs, Fine-tuning, Multimodal systems, Inference optimisation. Representative stack: OpenAI API, Anthropic Claude API, Google Gemini API, Hugging Face Transformers, vLLM, NVIDIA TensorRT-LLM, LiteLLM, Unsloth. 4. **Memory** (Context and memory systems): We give AI the continuity to understand the user, the work and the moment. Capabilities: Short-term context, User state, Persistent memory, Semantic memory, Retrieval, Context management. Representative stack: Mem0, Zep, Letta, LangMem, Cognee, Graphiti, Redis, Supermemory. 5. **Agency** (Agent harness and runtime development): We build the runtime that turns model capability into dependable execution. Capabilities: Planning, Execution, State machines, Checkpoints and routing, Retries and recovery, Reasoning workflows. Representative stack: LangGraph, OpenAI Agents SDK, Google Agent Development Kit, Pydantic AI, Mastra, Strands Agents SDK, Temporal, Restate. 6. **Orchestration** (Multi-agent orchestration): We orchestrate specialised agents around a shared outcome and accountable handoffs. Capabilities: Agent delegation, Supervisors, Handoffs, Parallel execution, Workflow coordination, Distributed systems. Representative stack: Microsoft AutoGen, CrewAI, Semantic Kernel, LlamaIndex AgentWorkflow, Agent2Agent (A2A), NVIDIA NeMo Agent Toolkit, Agno, CAMEL-AI. 7. **Action** (Tools, actions and integrations): We connect agents to the software, data and actions that businesses already depend on. Capabilities: APIs, Model Context Protocol (MCP), Browser automation, Databases, Enterprise software, Function calling. Representative stack: Model Context Protocol, Composio, Pipedream Connect, Browserbase, Playwright, Apify, Zapier MCP, Arcade. 8. **Evaluation** (Evaluation and observability): We make AI quality visible from development through production. Capabilities: Automated evals, Regression testing, Tracing, Task-success metrics, Hallucination detection, Production monitoring. Representative stack: Langfuse, Arize Phoenix, LangSmith, W&B Weave, Braintrust, TruLens, Ragas, OpenTelemetry. 9. **Governance** (Security, guardrails and governance): We engineer control, accountability and protection into every layer. Capabilities: Permissions, Sandboxing, Prompt-injection defence, Data protection, Auditability, Policy enforcement. Representative stack: NVIDIA NeMo Guardrails, Guardrails AI, Lakera, Microsoft PyRIT, Presidio, Open Policy Agent, HashiCorp Vault, Amazon Bedrock Guardrails. 10. **Scale** (AI infrastructure and economics): We optimise the infrastructure so intelligence stays fast, resilient and commercially viable. Capabilities: Deployment, Model serving, Latency, Caching, Token optimisation, Inference-cost engineering. Representative stack: Docker, Kubernetes, KServe, NVIDIA Triton, Ray Serve, BentoML, SkyPilot, Vercel AI Gateway. ## Domains - **Commerce and Finance**: Marketplaces, bookings, payments, fintech, and the trust and verification layer around every transaction. - **Digital Experience, Media and Growth**: Brand experiences, content and media platforms, streaming and publishing, community, engagement, and the growth systems that acquire and keep users. - **Health, Care and Wellbeing**: Care access, monitoring, coaching, fitness and wellbeing: systems people rely on for their health. - **Operations, Industry and Logistics**: Manufacturing, inventory, supply chains, fleets and field operations: the software that runs physical output. - **Physical and Connected Environments**: Property and the built environment, mobility, IoT and wearables, and safety and emergency response. ## Frontier R&D - **AI and ML systems**: Agent runtimes, memory, evaluation and inference economics: the open questions that decide whether AI holds up in production. - **Mixed Reality**: Spatial interfaces that carry intelligence into the physical world, where people work, train and move. ## How we work 1. **Frame**: Find the valuable decision, workflow or experience. 2. **Design**: Shape the interaction, context and control model. 3. **Engineer**: Build the intelligence, agent and tool chain. 4. **Prove**: Evaluate performance, safety and economics. 5. **Operate**: Deploy, observe and improve in the real workflow. Path: Opportunity → Prototype → Pilot → Production → Compounding. ## Case studies Asymmetri's work spans AI platforms, agent systems, SaaS products, mobile apps and original research for early-stage founders, fast-growing teams and global brands. Selected case studies include Prescribe Life, Komerz OS, Quicli, FinishLine and Project Protocol, across health, commerce, operations and AI research. ### [Prescribe Life](https://www.asymmetri.in/case-studies/prescribe-life): Coaching intelligence that works between sessions. AI platform · Web + mobile · Health, Care and Wellbeing · Practices: AI Agentic Systems and Data, Application Development, Innovation and R&D An AI performance-intelligence platform for leadership coaches. Behavioural science, biometric data and an AI co-pilot flag burnout risk between sessions. Challenge: Leadership coaches see a client for an hour at a time, but the signals that predict burnout show up in between: sleep, recovery, energy and mood. Prescribe Life needed a product that could read those signals and turn them into better coaching, without feeling clinical. Approach: PrescribeLife.AI is a performance intelligence platform for leadership coaches. It combines behavioural science, biometric data and AI to surface what is affecting a client's resilience, recovery and wellbeing between sessions. Its AI co-pilot, Preslie.AI™, tracks patterns, flags risks and delivers personalised micro-strategies, while coaches get a full 360° view of each client's goals, habits and readiness, so they can coach smarter and prevent burnout. Hard problem 1: Read a person's resilience from a phone, a watch and a sixty-second selfie. How do you measure recovery, stress and readiness across every iPhone, every Android and every wearable, including clients who own no wearable at all, and still give a coach a picture they can trust? - Frame: Two questions, not one dashboard. We sat with how coaches prepare for a session and found they ask two different things: how is this person's body doing, and how do they feel it's going? Most wellness products blend both into a single number. We framed the measurement as two lenses, objective and subjective, read side by side, with one rule on top: no client goes unmeasured just because they don't own a watch. - Design: The body on one side. The person on the other. The objective lens reads sleep, activity, readiness and wellbeing, each score explainable down to its factors: sleep debt, regularity, deep and REM sleep, resting heart rate, HRV. The subjective lens is a ten-item burnout and resilience instrument that places a client on a five-stage scale from Burnout to Flourishing, then ten resilience drivers the client rates against their own picture of what a ten looks like. Keeping the lenses apart means a gap between them becomes a signal, not an average. - Engineer: Wearable, phone and camera, in one pipeline. Apple HealthKit and Google Health Connect feed a health-data layer that collects in the background, merges devices and handles each platform's HRV measure. For everyone else there's the camera: a sixty-second remote photoplethysmography (rPPG) face scan reads heart rate, HRV, breathing rate and blood pressure from the tiny colour changes in the skin. Then the clever part: scan results are written back into Apple Health or Health Connect, so a selfie fills the same gaps a wearable would and every downstream score improves with it. - Prove: Refuse the impossible. Admit the unknown. Every camera reading is checked against physiological limits before it's stored: a heart rate outside 30 to 220, a breathing rate outside 5 to 60, a blood pressure no human could have, and the database rejects it. Missing sensors show as missing, never as zero. Change over a programme is only called improvement when the weekly average moves by more than a set threshold; anything smaller is reported, honestly, as holding steady. - Operate: Trends over moments. In the real workflow a single reading means little, so the product leans on trends: four-week health trends, month-on-month comparisons and averaged scan history, all clipped to the dates of each coaching programme. The onboarding flow for health permissions was rebuilt several times on real usage, until connecting a device became a step people finish rather than skip. Result: Two honest lenses, the body and the person, read side by side instead of blended into one opaque score. (60 s selfie scan to four vital signs; 2 health platforms, one pipeline; 5 stages, burnout to flourishing; 10 resilience drivers) Hard problem 2: Build an AI coach that knows the whole person, and knows its limits. How do you let an AI talk to high performers about stress every day, informed by their body and their mood, without it ever playing therapist or stepping between a client and their coach? - Frame: Not a chatbot. The bridge between sessions. We framed Preslie by what it is for: keeping the coaching going between sessions, and noticing what the coach would want to know. Then we wrote down what it must never do: diagnose, prescribe, pile pressure on someone who is already depleted, or keep a serious concern to itself. - Design: Tone follows the stage. Preslie reads where the client is and changes how it speaks. In Burnout it goes ultra-gentle, rest first, no plans. In Flourishing it celebrates briefly and helps sustain. The work splits into three orchestrated modes, conversation, behaviours and tips, each with agents for the reply, the suggested next replies and the actions that follow. Concerns are designed as a handoff: they become coaching topics for the human coach, and the client can hide a conversation while the coach still sees that something needs attention. - Engineer: The whole person in 150 tokens. A context agent compresses everything Preslie needs into about 150 tokens: the client's stage and score, sleep, activity, wellbeing and readiness, recent wins, focus drivers, today's behaviours and any open concerns. It refreshes the moment that data changes. Each job gets the right model: a frontier model for conversation, fast models for suggestions and insights, a tiny one for spotting wins. Every reply then passes a deterministic safe-language filter, so 'anxiety' becomes 'stress', 'you should' becomes 'you could', and nothing promises to treat or cure. - Prove: Precision over panic. A concern classifier that flags everything is as useless as one that flags nothing. Ours looks for specific signals, such as hopelessness, self-harm, severe insomnia or substance use, with explicit examples of what not to flag and a target of flagging only five to ten percent of messages. Low-confidence calls are dropped. Related messages over thirty days thread into one topic instead of ten alerts, and an urgent concern clears everything else off the screen. - Operate: Upgrade the brain without touching the body. Preslie has moved through several model generations since launch without rebuilding the app, because models are configuration, not code. Every AI call has a fallback, so the client always gets an answer. Daily insights are cached to keep costs flat as clients grow, and fixes ship as over-the-air patches. Coaches close the loop with programmes, weekly check-ins and reminders sent only to the clients who haven't checked in. Result: Preslie knows the whole picture, speaks gently, and hands the hard moments to a human. (−90% context per ai turn; 22 specialised agents; 5–10% target flag rate for concerns; 100% replies through the safe-language filter) Highlights: 360° coach portal (Clients, goals, habits, health snapshots and coaching topics on one screen, so coaches walk into every session already briefed); Preslie.AI™ co-pilot (Tracks patterns, flags risk and suggests personalised micro-strategies that a coach can assign as prompts or habits); Biometric and behavioural signals (Activity, sleep, readiness and wellbeing trends sit beside self-reported reflections and assessments); A resilience model (A burnout-to-flourishing scale turns assessments and vitals into a stage, a score and the drivers behind it); A habit loop on mobile (Clients choose priorities, track habits and chat with Preslie between sessions; wins and reflections flow back to the coach). Platforms: Coach web portal, Client mobile app. Services: Product design, Mobile app development, Web development, AI agent engineering, Rebranding. Tech stack: Mobile: Flutter, GetX, Shorebird; Health data: Apple HealthKit, Google Health Connect, Sahha, Shen.AI rPPG; AI: LangChain, OpenAI, Google Gemini; Backend: Supabase, PostgreSQL, Edge Functions, pg_cron, Firebase Cloud Messaging; Coach portal: Next.js. Outcomes: 60 s Selfie scan to vitals; −90% AI context per turn; 22 Specialised AI agents; 2 App stores, one codebase. ### [Komerz OS](https://www.asymmetri.in/case-studies/komerz-os): From a research question to a business plan you can sign. AI platform · Commercial planning · Commerce and Finance · Practices: AI Agentic Systems and Data, Application Development, Technology Consulting An AI commercial-planning platform for consumer brands. A research agent answers where to play, then a planning workspace builds how to win, down to a P&L that reconciles. Challenge: Komerz helps consumer brands decide where to play and how to win: which markets, channels, prices and packs, and what the P&L looks like if they get it right. The method lived in consultants' heads and spreadsheets. Komerz wanted it as software: a research partner that answers commercial questions, then turns the answer into a plan a brand can take to a buyer or a board. Approach: Komerz OS is two workspaces joined by one bridge. Where to Play is a research chat that reads the brand's own files, Komerz's proprietary market data and the public web, and answers with citations, reports, decks and workbooks. How to Win turns that research into channels, competitor pricing, price and pack, value chains, volume plans and a monthly P&L. We rebuilt the platform from the ground up on a serverless AWS stack, with an agent that orchestrates specialist models and a set of deterministic engines that own anything a finance team has to trust. Convert to project carries the whole conversation, its files and its template choices into the plan, so nothing gets typed twice. Hard problem 1: Make an AI build a P&L without inventing a single number. How do you let a model draft a three-year business plan when one plausible wrong margin can sink the deal it was built for? - Frame: The risk isn't a bad sentence. It's a believable number. We traced how a Komerz consultant actually builds a plan, from the shelf price down to what the manufacturer keeps, and marked every point where an AI could make a number up. That reframed the brief: the job wasn't to make the model better at spreadsheets, it was to take the arithmetic away from it entirely. - Design: Split judgment from arithmetic. We designed a control model with two owners. The model chooses: which products, which routes, which template fits and which scenario inputs to try, each with a stated reason and confidence. A compiler owns every formula. Edits follow the same split: you say what to change, the engine decides what that does to the plan. - Engineer: A compiler for commercial logic. Komerz's templates are written as row references, like R1 × R4. The compiler translates them into live Excel formulas, works out each row's role in the P&L from the waterfall, and refuses to build if net sales minus variable spend doesn't reconcile to manufacturer cost. The same formula engine runs in the agent and the API, held identical by a parity test, so a number in chat matches the number in the workspace and the download. - Prove: Audit every template, then reconcile to the last decimal. When one production run hit a circular reference, we didn't patch that template. We audited all of them. Three formula defects came out, including a value chain that ran backwards. Every template now compiles a full 8-sheet, three-year forecast as a test. Then we ran a live brand plan end to end and reconciled it against the source data. - Operate: Fast first, then right, then both. Research workbooks got the same treatment in production. One planning call designs typed columns, tabs are written in parallel, every row is mapped to its source, and the file is published first and verified after. Only the tabs that fail get repaired; passing tabs are kept byte for byte. Edits are routed by where a workbook came from, so a compiled plan never gets rewritten by a generic writer. Result: The model makes the calls. The compiler does the maths. (0.00 unallocated balance on a live plan; 2,704 formulas, zero errors; 9/9 templates audited and compiled; −66% time to first workbook) Hard problem 2: Let an agent think for twenty minutes on compute that quits at fifteen. How do you run deep, multi-step research in a multi-tenant product, where every turn is bounded, survives a crash and pays for each search exactly once? - Frame: The first version asked permission for everything. The original flow sent every message through a classifier with three buckets, then a keyword allowlist. It was predictable and it broke on the second follow-up. Watching real conversations showed what users wanted: an analyst who decides for itself when to search, when to stop and when to build something. - Design: Bounded agency. Our rule: the model decides whether a capability is needed; guarded tools decide how it runs. The agent gets about a dozen tools, such as inspect an attachment, search proprietary data, search public sources, check coverage, create an artifact. It gets no shell, no raw filesystem and no code execution. It works inside budgets it can see, and it can only create or change a file when the user's words grant it. - Engineer: Work that outlives the machine running it. The API accepts a prompt and returns immediately. The agent runs on its own and streams planning, thinking, task and answer events back over a real-time subscription. An idempotent lease ensures one prompt has one owner. Fingerprinted checkpoints on a per-chat file system keep every finished search and plan, so a worker that takes over picks up the evidence instead of paying for it again. Decks, workbooks and project builds run on their own workers and step functions, with dead-letter queues behind them. - Prove: Break it on purpose, and write down what broke. The first agentic build did worse than the old one: five of six runs crashed. We said so in writing, ran three independent review rounds and fixed what they reproduced. Every change since ships with a dated report: root cause from traces, the fix, the tests, and the same prompt run live against production for comparison. Transport tests fast-forward time to prove 18- and 50-minute turns still land. - Operate: Every bill itemised. Every change reversible. In production, each model call records its tokens, cached tokens, reasoning tokens and why it stopped, priced from a versioned catalogue across four providers and rolled up per prompt, user and month, with limits admins can set. Every major change ships behind a flag, from shadow to canary to default, with the previous path kept as an instant rollback. When a deck once took eighteen minutes, the trace showed a parser timeout being read as broken code; after the fix, the next live deck needed zero repairs. Result: The model decides what to do. The system decides how far it may go, and remembers where it stopped. (50 min longest turn proven in tests; 1× each paid search, even on retry; 1,874 automated tests; 18 agent eval scenarios) Highlights: Where to Play research chat (Ask a commercial question in plain language, attach your own files, and get a cited answer with the research activity visible as it runs); Decks and workbooks on demand (Reports, presentations with live charts and Excel workbooks with real formulas, each editable in place when you ask for a fix); Convert to project (A useful chat becomes a How to Win project, with the transcript, files and template choices carried across and the intake pre-filled); Market-specific channels (Routes come from each market's own channel architecture, grouped by in-home, out-of-home, on-the-go, travel retail and omnichannel); Value chain to P&L (Template-driven value chains feed customer and channel volume plans, which roll up into a monthly P&L and a verified business-plan workbook); Evidence you can check (Sources, assumptions and validation gaps stay visible. Missing data shows up as a need, never as a silent zero). Platforms: Web workspace, Research agent, Serverless platform. Services: Technology consulting, AI agent engineering, Product design, Platform engineering. Tech stack: Frontend: Next.js, React, Vercel AI SDK, Univer, ExcelJS; API: AWS AppSync, GraphQL, Stytch B2B; Agent: Python, deepagents, LangChain, AWS Lambda, AWS Step Functions; Models: Claude on Amazon Bedrock, OpenAI, Google Gemini; Data and search: DynamoDB, Amazon S3 Files, Neo4j, Exa; Operations: AWS CDK, AWS Amplify, Langfuse, Amazon SQS. Outcomes: −66% Time to first workbook; 0.00 Unallocated P&L balance; 1,874 Automated tests; 5 mo First commit to V4. ### [Quicli](https://www.asymmetri.in/case-studies/quicli): From 'Hi' on WhatsApp to a doctor's prescription in minutes. Health platform · WhatsApp + web + doctor app · Health, Care and Wellbeing · Practices: AI Agentic Systems and Data, Application Development Doctor consultations on WhatsApp. Say 'Hi', describe your symptoms in your own language, and a real doctor's prescription arrives in the same chat, with medicines and tests ready to order. Challenge: In India, the first doctor most people consult is a search engine, and one search for a symptom can turn a cough into a night of worry. Seeing a real doctor means an appointment, a queue, a fee and often a video call nobody asked for. Quicli wanted to make proper medical advice as easy as sending a message, for everyone: the office worker between meetings, the parent with a feverish child, the retired veteran who finds apps confusing. Approach: Quicli is a symptom-to-resolution service that starts on WhatsApp. A patient sends 'Hi', creates a profile for themselves or a family member, and answers a few simple questions about their symptoms in their own language. Their answers become a structured clinical summary for a real doctor, who reviews it and sends a signed prescription back to the same chat. Prescribed medicines and lab tests are added to a cart automatically, the prescription can be read in the patient's own language, and Quicli checks in when the course of medicine ends. We designed and built the patient experience, the doctor app and the backend that connects them. Hard problem 1: Take a doctor's history from someone who will never download an app. How do you take a proper clinical history from a worried parent, a 64-year-old veteran or a first-time smartphone user, in their own language, without asking them to install anything or type a single medical word? - Frame: The competitor is a search engine at 2am. Quicli's rival isn't another health app. It's the late-night symptom search that leaves people more scared than informed. We framed the product around the people most failed by apps and appointments, and around one rule for the doctor: the case must arrive structured, not as a chat transcript to untangle. - Design: Chat for the conversation. One tap for the questions. We split the job by what each surface does best. WhatsApp handles the front door, status updates, the prescription and the follow-up. A light web flow, opened from a single button in the chat, handles everything WhatsApp can't: a branching medical questionnaire, pictures of body regions to tap, voice notes, photos and payment. Questions are multiple choice; patients pick what they speak and what they understand separately, because those are often different languages. - Engineer: Thirteen thousand paths, and none of them dead-ends. Behind the simple questions sits a clinical decision tree: 23 conditions, 138 questions and 13,216 answer combinations, each mapped to a likely diagnosis, medicines, tests and advice, and each flagged if it needs a hospital visit or a video call. Every question exists in 11 languages with three phrasings, so it reads naturally. Any option that would lead to an uncovered combination is hidden before the patient sees it. An AI then writes a short clinical summary for the doctor, with a mandatory red-flag line for symptoms that need urgent care. - Prove: Audit every path before a patient walks it. A coverage audit walks the whole tree and lists every answer combination without a diagnosis. Condition by condition, those gaps were closed. Small-phone details that make or break usability, like inputs that zoom the page and break the flow, were found and fixed. And whatever the AI writes, a doctor reviews the case and signs the prescription. - Operate: A prescription people can actually read. The signed prescription arrives in the chat as a PDF, with an offer to read it in the patient's own language. Translation keeps drug names in English for safety and turns dosing like 'before breakfast' into the patient's language, rendered with fonts for nine scripts and cached per language. Ratings are three buttons; an unhappy rating turns the patient's next message into the reason. When the medicine course ends, an approved WhatsApp template checks in, with fallbacks if the message window has closed. Result: Chat for the conversation, one tap for the questions, and the patient's own language from 'Hi' to prescription. (0 apps to download; 11 languages, end to end; 13,216 answer paths, no dead ends; 23 conditions covered) Hard problem 2: Get a real doctor to answer in ten minutes, for ₹99. How do you promise a signed prescription within ten minutes at a price almost anyone can pay, when every case needs a qualified doctor's attention and doctors are only human? - Frame: The doctor's time is the product. We worked back from the promise. Ten minutes from report to prescription, at a price that has to cover a doctor, means the doctor should spend those minutes deciding, not asking routine questions or searching for drug names. The case has to arrive ready, and the routing has to make waiting impossible. - Design: A case that arrives already worked up. The doctor app opens on a prepared case: the structured summary, the patient's history and past consultations, likely diagnoses and suggested medicines and tests from the decision tree. Voice notes are transcribed. Video is offered only when the law requires it or the doctor wants it. Doctors book two-hour shifts, and each shift has a capacity, so the load is planned rather than hoped for. - Engineer: No case waits on one unread alert. When a case is assigned, the doctor gets a high-priority alert every 30 seconds. If they haven't accepted within two minutes, the case moves automatically to the on-shift doctor with the most free capacity who hasn't already been tried, and the patient is told on WhatsApp. Assignment and capacity update together in one transaction, so two doctors never get the same patient. Prescribing searches a catalogue of about 254,000 Indian medicines, and the doctor signs on screen. - Prove: Test the chain end to end. The alert and reroute chain is covered by automated tests and was verified end to end with shortened timers before release. Real edge cases shaped it: consultations created already assigned once skipped the first alert, and that was fixed. Doctor ratings and a patient feedback form close the loop on quality. - Operate: From prescription to medicine cabinet. Doctors move on and off duty automatically with their shifts. Once a prescription is signed, the medicines and lab tests are already in the patient's cart, with checkout and delivery a tap away. And because Quicli remembers past consultations, the next doctor starts with the patient's history instead of a blank page. Result: Do the doctor's paperwork before they open the case, and never let a patient wait on one unread alert. (10 min prescription promise; 30 s doctor alert cadence; 2 min automatic reroute; ₹99 per consultation with a pass) Highlights: Starts with 'Hi' (Scan a QR code or message the number; no app download, no account to set up first); Family profiles (Several patient profiles on one WhatsApp number, each with its own history); Symptoms in your language (Multiple-choice questions with body-part pictures and a voice-note option, in 11 languages); A doctor, automatically (Each case is routed to an on-shift doctor, with a structured summary ready to review); Prescription in the chat (A signed prescription PDF on WhatsApp, available in the patient's own language); Medicines and tests, carted (What the doctor prescribes is already in the cart, ready to order in one tap). Platforms: WhatsApp, Patient web app, Doctor app (iOS, Android, web). Services: Product design, Web development, Mobile app development, AI engineering. Tech stack: Patient: WhatsApp Cloud API, WhatsApp Flows, Next.js, React, Tailwind CSS; Doctor app: Flutter, GetX, VideoSDK, Firebase Cloud Messaging; Backend: Firebase, Cloud Functions, Firestore, Cloud Tasks; AI: Open-weight LLMs on Groq, Google Gemini, OpenAI audio transcription. Outcomes: 0 Apps to download; 11 Languages; 13,216 Answer paths, no dead ends; 2 min Automatic doctor reroute. ### [FinishLine](https://www.asymmetri.in/case-studies/finishline): One place to run the studio, for people and their agents. Internal platform · Web app + AI agents · Operations, Industry and Logistics · Practices: AI Agentic Systems and Data, Application Development Asymmetri's own delivery operating system: projects, QA, clients, chat and calendars in one secure app, where AI agents work as the person who sent them. Challenge: A studio that ships for clients runs on a dozen tools: one for tasks, one for QA, chat somewhere else, calendars, leave and meeting notes in others, and clients asking for updates by email. Every hand-off between them loses context, and the newest teammates, AI agents, couldn't safely touch any of it. So we built our own. Approach: FinishLine is Asymmetri's delivery operating system. Projects, tasks, boards, timelines and sprints sit next to multi-round QA, a client portal, realtime chat, Google Calendar, meeting notes, leave, learning and each person's plan for the day, with two-way Slack, GitHub and Linear sync for the work that lives elsewhere. Every surface (the web app, the client portal, background jobs, inbound integrations and AI agents) goes through the same row-level security in Postgres across five roles. Claude, ChatGPT and Cursor connect through a hosted MCP server and act as the person who sent them, with the same permissions, notifications and audit trail. Hard problem 1: Make an app that checks every permission on every row feel instant. How do you keep security in the database, five roles deep, and still open a project of 850 tasks before anyone notices the wait, on serverless infrastructure that goes to sleep? - Frame: Measure before you guess. We profiled where a page actually spent its time and found it wasn't where anyone expected. The home screen alone made about 1,450 project-access checks per load. Query planning cost more than some queries. Every page waited on a network round trip just to confirm who was logged in, and the first visitor after a quiet spell paid for the cold start. - Design: Never trade security for speed. Change how it's evaluated. The rule stayed: the database is the only security boundary. What changed was the shape of the work. Instead of asking 'can this person see this row?' 850 times, ask once 'which rows can this person see?' and let Postgres join against the answer. Verify sessions locally. Defer everything the user isn't waiting for. For realtime, send ids, not rows, and let each browser fetch what it's allowed to see. - Engineer: Shave every layer. Permission policies were rewritten as hashed sets, hot functions pinned to generic plans, and read and write policies split so reads stop planning edit joins. Sessions are verified with signed tokens in the function itself. Heavy libraries load only when needed, the browser database client is lazy-loaded, and notifications are sent after the response returns. Boards and QA statuses update optimistically and roll back on failure. - Prove: Faster, and provably just as strict. Speed means nothing if it leaks. Every rewrite ran against a permission test suite that signs in as each role and asserts what it may and may not see: 1,666 checks for the task rewrite alone, and about 3,650 across the product. The lean home query was proven to return exactly what the full one did for seven roles. Timings are medians of repeated cold runs, not a lucky best case. - Operate: Keep it warm. Keep it honest. A scheduled ping keeps the app warm so nobody waits for a cold start, at about 4% of the hosting free tier. Live updates stay well inside the realtime connection limit. When one bundling change broke production, it was rolled back the same day and redone the safe way, and the commit says so. Result: Security stayed in the database. We made the database fast. (−66% task read time for a manager; 4× faster server load; −36% 300-row list render; 3,650 permission assertions in tests) Hard problem 2: Let AI agents do real work, as exactly the person who sent them. How do you let Claude, ChatGPT or Cursor read and change projects, tasks and QA without opening a side door around the permissions, approvals and notifications the team relies on? - Frame: Agents are teammates, not integrations. We reframed the brief. The question wasn't 'what should the API expose?' It was 'what should a teammate who happens to be an agent be allowed to do?' The answer: whatever the person who sent it can do, nothing more, and in plain sight. - Design: Same door as a human. Agents don't get their own endpoints. A write from an agent builds the same request the web app would send and runs the app's own action, signed in as that person. Validation, permissions, notifications, Slack copies and emails are identical by construction. Each person chooses Off, Read, or Read and edit for their agents, admins can block individuals, and the company has a single switch. - Engineer: A hosted MCP server with its own OAuth. FinishLine is its own OAuth 2.1 authorization server, with discovery, dynamic client registration, PKCE, revocation and rotating refresh tokens that detect replay. On approval it creates a real session for the user, stores it encrypted, and keeps only hashes of tokens. Thirty-three tools cover the studio: fifteen to read, fifteen to act, plus search, identity and a confirmed delete. - Prove: The agent hits the same wall a person does. Because agents go through the same door, the same permission suite covers them, with dedicated tests for agent access and live end-to-end tests where an agent reads and writes real records. Rules that must stay human, like learning results, are guarded in the database itself, so no prompt can talk its way past them. - Operate: Every agent action, in plain sight. Each agent action is logged, and notifications say who sent it: 'via Claude'. Rate limits and a company-wide kill switch sit on top. The studio now plans its day, files tasks, updates QA and writes notes through agents, and the team built FinishLine's second version the same way, with AI-assisted engineering through most of its 674 commits. Result: An agent in FinishLine is its user: same door, same rules, every action on the record. (33 agent tools over mcp; 1 rule set for people and agents; 3 agent clients supported; 0 side doors around permissions) Highlights: Projects four ways (List, board, timeline and sprints over the same tasks, with drag-and-drop that updates instantly); QA with a two-person rule (Multi-tester rounds, fix rounds and client acceptance, and nothing is done until someone who didn't build it has passed it); A client portal (Clients see progress, approve QA, raise requests and chat, with their own Slack bridged in); Calendars and meetings (Google Calendar push sync, meeting notes whose action items become tasks, and leave with approval chains); Two-way Linear and GitHub (Issues, pull requests and checks stay in step with the work, without echoes or lost updates); AI agents as teammates (A hosted MCP server lets Claude, ChatGPT and Cursor read and act in FinishLine as the person who connected them). Platforms: Web app, Installable PWA, Client portal, MCP server. Services: Product design, Web development, AI agent engineering, Platform engineering. Tech stack: App: React, React Router, Mantine, TipTap, dnd-kit, Vite; Data: Supabase, PostgreSQL, Row-level security, Supabase Realtime, pg_cron; Hosting: Vercel, Service worker / PWA, Web Push; AI: Model Context Protocol, OAuth 2.1; Integrations: Google Calendar, Slack, GitHub, Linear, Fathom, Resend. Outcomes: −66% Task read time; 4× Faster server load; ~3,650 Permission assertions; 33 AI agent tools. ### [Project Protocol](https://www.asymmetri.in/case-studies/project-protocol): What if machine knowledge could compound outside the model? Research · Multi-agent AI · AI and ML systems · Practices: AI Agentic Systems and Data, Technology Consulting, Innovation and R&D Asymmetri research into shared machine cognition: can teams of AI agents write what they learn into a shared, inspectable layer that the next agents inherit and build on? Challenge: Every AI agent starts its working life from zero. It may be brilliant, but what it figures out on Tuesday is gone by Wednesday unless someone retrains the model, which is slow, costly and tied to one vendor, or stuffs old transcripts back into the prompt. Human intelligence didn't scale that way. It compounded because meaning moved out of individual minds into language, writing and institutions that the next generation inherited. Nobody has built that for machines. Approach: Project Protocol is Asymmetri's research programme into shared machine cognition. The hypothesis: a team of AI agents can externalise useful structure (concepts, situation patterns, strategies, failure corrections and their lineage) into a shared protocol layer outside any model's weights, and later agents can inherit that layer and do better than agents with no memory or with plain transcript search. We built a deterministic simulation and experiment harness to test it honestly: a world that rewards abstraction, a typed and auditable protocol layer, LLM agents that read it through tools, and baselines strong enough that a positive result would mean something. Then we turned the same rigour on our own results. Hard problem 1: Make machine knowledge compound outside the model's weights. Can a team of AI agents write what they learn into a shared, inspectable layer that the next generation inherits and builds on, the way human knowledge outlived every individual mind? - Frame: Language, writing, institutions. Then machines. We started from a thesis about how intelligence scales. Humans didn't get smarter by growing bigger brains every generation; they moved meaning into shared forms that outlast any one mind. We framed the machine equivalent precisely, including what it is not: not a vector database, not a memory log, not an API contract, not a central orchestrator. A protocol layer of concepts, schemas, reasoning patterns, failure corrections and lineage. - Design: Build a world where abstraction pays. An idea like this needs a world that rewards it. We designed a supply grid split into four sectors, one agent each, with three resources, demands on deadlines and corridors that choke. Each agent sees only its own sector clearly and must coordinate across boundaries on a message budget set at half of what it would naturally use. The layer itself holds typed situation patterns and strategies, each carrying evidence of when it helped or hurt, retrieved by a structural vocabulary rather than embeddings so every success and failure can be inspected. - Engineer: A deterministic lab for agents. We built a deterministic simulation and experiment harness so that every comparison is exact and repeatable. Agents run with no memory, with the protocol layer, with transcript search at the same read budget, or with full sight of the world as an oracle ceiling. LLM agents read the layer through tools and propose moves that a validator checks, running first in shadow mode beside a scripted agent and then live. - Prove: Find the headroom before chasing the gain. About 150 logged episodes later, the first thing the lab proved is that the prize is real: an agent that could see everything fulfils nearly twice the demand of today's agents, so there is a large gap for shared knowledge to close. The first signals of the layer helping appeared exactly where coordination is hardest, on cascading chokepoints. They are early and small, and we treat them that way. - Operate: Run it like a lab, not a launch. Every round of runs ends in a written learnings report: what the data showed, what it could not support, and what changes next. Those reports reshaped the agents, the world and the question itself, and pointed straight at the next, sharper experiment. Result: If knowledge can compound outside the weights, every fleet of agents gets an institutional memory. (1.9× headroom: oracle vs today's agents; ~150 logged episodes; 4 conditions, incl. an oracle ceiling; 180 token read budget per agent) Hard problem 2: Build an experiment that can't fool its own builders. How do you test a big idea about machine intelligence so that a 'yes' really means yes, when every shortcut makes a yes easy to manufacture? - Frame: Our best early result was wrong. We said so. The first runs showed the protocol layer cutting agent communication by about three quarters. It would have made a great headline. We took it apart instead and found the cause: the protocol condition used shorter message templates. We normalised them, the advantage mostly vanished, and we wrote it down. That reframed the programme: the hard problem isn't building the layer. It's proving it. - Design: A world no model has seen before. The next phase, GLYPHWORLD-R, generates universes with randomised physics and nonsense vocabulary, so no model can lean on what it already knows. Some rules are never observable at all: only an agent that has formed a real abstraction can get them right. A cosmetic decoy catches superstitions, and three roles each see part of the physics, so the full picture only exists if they build it together. - Engineer: Thirteen ways to be wrong, tested side by side. Instead of one comparison, the design runs a full set of arms that isolate each part of the idea: no memory, transcript search, search over the layer's own history, append-only notes, a curated shared layer that agents can add to, support, contradict, revise and merge, private layers per role, a placebo, a perfect rulebook, a one-shot summary, a reader from a different model family, alternating writers, and lesions that remove parts of the layer to see what breaks. - Prove: Decide what counts before you look. The statistics are pre-registered: 20 to 28 universes, paired tests with bootstrap confidence, correction for multiple comparisons, equivalence tests so a null can be a real answer, and an early futility check. The analysis code is validated on synthetic data before any real run. The design itself won a unanimous vote of a three-judge review council against four competing proposals. - Operate: Frontier questions on a startup budget. The definitive run is costed at roughly $210 to $280 of model calls, with a hard cap of $350 and about four weeks of work. Fast, cheap, falsifiable: the kind of research loop that lets a small team ask a frontier question and get a real answer. Result: The first thing a research programme should get right is how it will know it's wrong. (13 experimental arms; 6 flaws in our own work, designed out; 1 headline result we retracted; $350 hard cap for the definitive run) Highlights: A protocol layer (Typed patterns and strategies with usage evidence, lineage and a lifecycle from proposed to retired); A world that rewards abstraction (A multi-sector supply grid with partial sight, deadlines and a scarce message budget); Fair baselines (No memory, transcript search at an equal read budget, and an all-seeing oracle ceiling); Pre-registered next phase (GLYPHWORLD-R: randomised physics, held-out rules and thirteen experimental arms). Platforms: Simulation harness, Experiment framework, LLM agent runtime. Services: Research design, AI agent engineering, Experiment design, Evaluation. Tech stack: Research harness: Python, Deterministic simulation, unittest; Agents: OpenAI Responses API, Tool calling, Validator-checked actions; Methods: Pre-registered statistics, Paired bootstrap, Equivalence testing, Ablation and lesion arms. ## Careers & Culture Asymmetri is a remote-first, async-first technology company based in Kerala, India. Its flat team of engineers, designers and applied AI specialists solves hard problems for global brands across four practices: AI agentic systems and data, application development, technology consulting, and innovation and R&D. Culture code: - **Flat by design.** No bosses, just ownership. Everyone's voice counts, including yours in week one. - **Feedback is fuel.** Direct but kind. Give it, get it, use it to get better. It's never personal; it's how we grow. - **Async-first.** We work across time zones and schedules. Write it down, say it clearly, and nobody waits on a meeting. - **Done ≠ perfect.** Push early, iterate fast. Shipped and improving beats perfect and imaginary. - **Decisions over indecision.** We'd rather you try, learn and adjust than stay stuck. A call made today beats a perfect one made next month. - **Need help? Ask.** No one expects you to know it all. Hitting friction? Loop people in early. Ask early, ask often. - **Everything is figureoutable.** New domain, unfamiliar stack, no template. We pull the problem apart until the answer shows itself. - **Celebrate everything.** Big wins and small ones. A win is a win, and we say so out loud. How a week works: - **Working hours**: Flexible. What matters is output, not hours online. Tell us your plans and we'll work around them. - **Meetings**: Fewer, faster, purposeful. If it can be a message, it stays async. - **Non-negotiable**: Daily morning syncs. A quick sync every morning to align on what's on everyone's plate. That's the one fixed point. - **Feedback**: Direct but kind. In the open, as the work happens. Not saved up for a review cycle. Open roles: https://www.asymmetri.in/careers#open-roles ## Contact To contact Asymmetri, email hey@asymmetri.in or send a brief through the form at asymmetri.in/contact-us. Asymmetri is based in Kerala, India and works with global brands worldwide. New projects start with a framing conversation, followed by a proposal for a prototype or pilot. 1. **You send the brief**: A few lines on the problem, the stakes and the timeline. No deck, no RFP required. 2. **It lands with the team**: Your note goes to the people who would build the answer, not into a ticket queue. 3. **A framing conversation**: We find where AI or new technology creates real operating leverage, and where it doesn't. 4. **A proposal**: A clear plan for a prototype or pilot: scope, team, timeline and what success looks like. ## FAQ ### What is Asymmetri? Asymmetri is a hard problems technology company. It engineers AI systems, agents and digital products for global brands, across four areas: intelligence, process, experience and scale. Asymmetri works through four practices: AI agentic systems and data, application development, technology consulting, and innovation and R&D. ### What does a hard problems technology company do? A hard problems technology company takes on consequential problems that off-the-shelf tools cannot solve: workflows where AI has to decide and act, products people depend on every day, and systems that must work in the physical world. Asymmetri invents the approach, proves it, then makes it repeatable. ### What AI services does Asymmetri offer? Asymmetri engineers the full AI stack as ten connected layers: AI product and interaction design, data and knowledge architecture (RAG, vector search, knowledge graphs), model and inference engineering, context and memory, agent runtimes, multi-agent orchestration, tool integrations including MCP, evaluation and observability, security and governance, and AI infrastructure economics. ### How is Asymmetri different from an AI consultancy or a software agency? Asymmetri does not add AI around the edges or stop at demos. Strategy, design and engineering stay in one accountable loop, from framing the opportunity to operating the system in production, with evaluation, governance and unit economics built in from the start. ### Which industries does Asymmetri work in? Asymmetri is domain agnostic. The team starts from the hard problem, not the industry, and brings the same approach to any sector. Recent work spans commerce and finance; digital experience, media and growth; health, care and wellbeing; operations, industry and logistics; and physical and connected environments such as property, mobility, IoT and wearables. ### Which AI models and frameworks does Asymmetri use? Asymmetri is model-agnostic and lets the use case decide the architecture. Teams choose between OpenAI, Anthropic Claude, Google Gemini and open models served with vLLM, and build with tools such as LangGraph, the OpenAI Agents SDK, Model Context Protocol, Langfuse and Kubernetes. These are representative technologies, not a fixed stack. ### How does an engagement with Asymmetri work? Every engagement follows one loop: Frame, Design, Engineer, Prove and Operate. Asymmetri finds the valuable decision or workflow, shapes the interaction and control model, builds the intelligence and tool chain, evaluates performance, safety and cost, then deploys and improves it in the real workflow, moving from prototype to pilot to production. ### How do I start a project with Asymmetri? Send a short note about the problem through the contact page or email hey@asymmetri.in. The first step is a framing conversation to find where AI or new technology creates real operating leverage, followed by a proposal for a prototype or pilot. ### What is it like to work at Asymmetri? Asymmetri is flat, remote and async-first. There are no bosses, just ownership; feedback is direct but kind; meetings are few and purposeful, with one short sync every morning. Most people work on at least three projects across different domains at once, close to the founders and with real input into what gets built. ### Is Asymmetri a remote company? Yes. Asymmetri is remote-first and async-first, and the team works across time zones and schedules. Hours are flexible: what matters is output, not face time, as long as timelines are met and the team stays in the loop. ### What roles does Asymmetri hire for? Asymmetri hires across engineering, design, applied AI and data, operations and QA, and growth. Open positions, including full-time, part-time and internship roles, are listed on the careers page at asymmetri.in/careers. ### How do I apply for a job at Asymmetri? Choose a role from the open roles on the careers page at asymmetri.in/careers and apply with the form on that role's page, including your resume. If nothing fits, email hey@asymmetri.in with what you would build at Asymmetri. ### What kind of work will I do at Asymmetri? Work spans Asymmetri's four practices: AI agentic systems and data, application development, technology consulting, and innovation and R&D. Projects run across domains such as commerce and finance, health, media, logistics and connected environments, so expect agents, products, platforms and prototypes for global brands. ### What tools does the Asymmetri team use? The team works AI-native. AI-assisted tools such as Claude Code, Codex, Cursor, Claude for design, Figma AI sit alongside continuous delivery, so engineers and designers spend their time on judgment rather than boilerplate. ### How do I contact Asymmetri? Email hey@asymmetri.in or send a brief through the form at asymmetri.in/contact-us. Include the problem you want solved, your timeline and how to reach you. ### What should I include in my first message to Asymmetri? A few lines are enough: what you are trying to achieve, what is in the way, where the work stands today (opportunity, prototype, pilot or production) and any deadline. No deck or formal RFP is needed to start. ### What happens after I contact Asymmetri? The team reads your note and replies to set up a framing conversation, which finds where AI or new technology creates real operating leverage. That is followed by a proposal for a prototype or pilot, then delivery through one loop: Frame, Design, Engineer, Prove and Operate. ### Does Asymmetri work with companies outside India? Yes. Asymmetri is based in Kerala, India and works with global brands worldwide, collaborating remotely and asynchronously across time zones. ### What kinds of projects does Asymmetri take on? Hard problems across four practices: AI agentic systems and data, application development, technology consulting, and innovation and R&D, in domains such as commerce and finance, health and wellbeing, media, logistics and connected environments. ### How do I apply for a job at Asymmetri? Job applications go through the careers page rather than the contact form. Open roles are listed on the careers page at asymmetri.in/careers, and each role page has its own application form. ## Pages - [Home](https://www.asymmetri.in/): positioning, AI system, domains, process and FAQ - [Case studies](https://www.asymmetri.in/case-studies): each built around the hard problems solved, through the engagement loop, with screens and build details - [Careers & Culture](https://www.asymmetri.in/careers): culture code, how we work, open roles - [Open roles](https://www.asymmetri.in/careers#open-roles): remote jobs at Asymmetri - [Contact](https://www.asymmetri.in/contact-us): send a brief or email hey@asymmetri.in - [AI Agentic Systems and Data](https://www.asymmetri.in/what-we-do/ai-agentic-systems-and-data): Systems that think, decide and act on your data. - [Application Development](https://www.asymmetri.in/what-we-do/application-development): Products people transact with, depend on and trust. - [Technology Consulting](https://www.asymmetri.in/what-we-do/technology-consulting): Frame the hard problem. Architect the answer. - [Innovation and R&D](https://www.asymmetri.in/what-we-do/innovation-and-rd): What is next, engineered into what ships.