AI Engineer (Typescript)

Kindgeek is looking for an AI Engineer who treats model output as a hypothesis, not an answer. We’re looking for someone who can build multi-agent systems and still clearly understand where a probabilistic call must stop and a deterministic rule must take over. Your mission will be to design and run the AI layer of a production fintech platform for our US-based client , where agents extract and reconcile data from real, messy documents — and every number they touch is eventually checked against a rule that never guesses. You’ll work in a small, senior team with direct ownership from architecture to production behavior. Responsibilities As an AI Engineer (TypeScript) , you will be responsible for: Designing multi-agent extraction and orchestration pipelines where each agent retrieves and uses only the context relevant to its task, rather than relying on naive keyword or similarity search. Building structured extraction, classification, and drafting features with output validation and clear boundaries against deterministic business logic. Integrating in-house ML prediction and detection endpoints into product features, handling timeouts and low confidence without fabricating a value. Making AI and ML behavior measurable through tracing, evaluation, and prompt/agent iteration, so quality is observed, not assumed. Designing durable, event-driven workflows around LLM and ML calls that handle retries, partial failures, and idempotency. Working with real, messy documents as first-class input — parsing, chunking, and structuring mixed financial and unstructured content. Writing production TypeScript across the backend, with typed contracts between AI services and the rest of the system. Requirements We’re looking for an AI Engineer (TypeScript) who has: Production experience designing or operating multi-agent AI systems , including how agents store, catalog, and hand off extracted context to each other. Practical experience shipping at least one LLM-powered feature to production , and the ability to describe a specific failure mode you found and how you contained it. Hands-on daily use of AI coding tools such as Claude Code, Codex, or equivalent, with clear judgment about what to verify before trusting their output. Strong production TypeScript and Node.js experience, comfortable with ESM and typed API contracts such as tRPC, GraphQL, or equivalent. Experience consuming ML predictions in application code , understanding confidence and uncertainty well enough to decide what the product should do when a model is wrong or unavailable. Experience with asynchronous, event-driven backend work — job queues or durable workflow engines — and the operational instincts that come from running them. Working knowledge of PostgreSQL or another relational database in production, including schema and migration basics. The ability to walk through a real technical trade-off you made and explain the reasoning behind it, rather than just…