AI Architect for the next generation of scalable AI systems, in the Founder & CEO’s office.

Own the system design.

You own how production AI is designed here: the agentic orchestration, the retrieval and data layer, the evaluation and guardrails, and the inference and cost architecture that keeps it running at scale. You set the patterns and standards, make the build-vs-buy calls, and stay hands-on in the code. This is a system-design role, not prompt engineering.

The work.

• Own the zero-to-one architecture of our core AI infrastructure, making the build-vs-buy calls on models, vector stores and orchestration.
• Architect autonomous, multi-agent workflows: agents that plan, call tools, write and run code, retrieve knowledge, and recover from their own errors.
• Design scalable, high-fidelity RAG: chunking, embeddings, hybrid search, re-ranking and evaluation, tuned for near-zero hallucination at scale.
• Run LLMOps end to end: observability, prompt and token control, semantic caching, and routing simpler tasks to small or fine-tuned models to control cost and latency at scale.
• Stay hands-on: write production Python, ship FastAPI services, manage Docker and Kubernetes, and later hire and lead a small team of elite engineers.

This part matters most.

• A systems architect. You have designed RAG and agentic systems end to end: hybrid search and re-ranking, GraphRAG, evaluation harnesses (RAGAS, DeepEval), guardrails and prompt-injection defense, orchestration with LangGraph and CrewAI, and LLMOps observability (Langfuse, Arize) across AWS, GCP or Azure, including Bedrock and Vertex AI.
• Agentic-first. Tool use, memory, multi-step reasoning and autonomous decisions in real workloads, with robust state and error handling.
• A builder. You live in the codebase and would rather ship a prototype over the weekend than debate theory for three weeks.
• Entrepreneur mindset. You obsess over unit economics and ROI, never over-engineer, and are resilient enough to scrap and pivot overnight. Bonus if you have started your own company, even if it failed.
• 3 to 4 years of high-density experience. Not a fresher role. You have taken generative AI from prototype to production and shipped systems real users depend on.

Show us your architecture.

Please note: personal projects, portfolios and profiles are not accepted. To be considered, you must complete one of the two cases below.

Instead of a whiteboard interview, we ask you to build. Not the whole product, just one meaningful, production-minded slice of the AI infrastructure for an AI-native B2E (business-to-employee) travel and expense platform, where employees ask in plain language and the system must act inside company policy and stay cheap to run at scale.

Pick one and build a working vertical slice. Both are demanding by design:

• Policy-aware trip-planning agent. Given “Book BLR to DEL next Tuesday, return Thursday, under policy,” it retrieves the travel policy (RAG), calls mocked flight and hotel tools, proposes a compliant itinerary, and flags anything out of policy with its reasoning.
• Expense-automation agent. Given a receipt plus the policy, it extracts the fields, categorizes the spend, checks compliance, flags anomalies or duplicates, and outputs an audit-ready record.

Your build should show: your retrieval design and how you keep the agent from hallucinating policy; multi-step agent orchestration, tool use and recovery from failed or partial tool calls (mocked APIs are fine); at least one eval that catches a wrong or out-of-policy answer; and how you control token cost and latency at production scale.

Deliverable: a GitHub repo plus a 5 to 8 minute walk-through video (a GDrive link or a Loom) covering your architecture, your build-vs-buy calls and what you would do next. Aim for a weekend build; we judge architecture and judgment, not completeness.

Details.