Senior AI Full-Stack Engineer

What we are looking for

Senior AI Full-Stack Engineer

Company: Grey Chain AI

Location: Remote

Experience: 8+ Years

Employment Type: Full Time

About the Role

We are growing our engineering teams and hiring senior full-stack engineers who have already shipped LLM-powered features to real users and want to do it at the level of a platform rather than a single app. You will work flexibly across two kinds of work, and most engineers here do both. Platform work: building the capabilities of our agent platform itself new agent types, tools and MCP servers, skills, the codeexecution sandbox, workflow and long-running task infrastructure, ingestion pipelines, and the graph, vector and ontology layers that agents reason over. Product work: building client-facing AI products on that platform, from data model to API to UI, integrated into the client’s Microsoft 365, Salesforce or identity estate, and deployed into their own cloud environment. Both kinds of work share a shape: a Python backend, a React/Next.js front-end, PostgreSQL underneath, an LLM in the loop, and an enterprise client at the end of it who expects the result to be secure, multi-tenant, auditable and fast. You will own features end to end schema, service, agent behaviour, UI, tests, and the evals that prove the LLM part actually works. You will join engineering teams that already ship to production. We expect you to raise the bar on them: through the quality of your own work, your code reviews, and your willingness to say when something is not good enough.

What you will build and own

Agentic features, end to end. Design and build agent workflows that solve real enterprise tasks: retrieval over client documents and structured data, multi-step tool use, specialised agents that hand off to one another, long-running jobs with durable state, and human-in-the-loop checkpoints. Own the prompts, the tool definitions, the orchestration logic and the evals not just the API around them.

Tools, MCP servers and skills. Build the tools agents call MCP servers over client systems and internal services, skills that package repeatable capabilities, and sandboxed code execution for data analysis. Design them so they are safe to expose to an LLM: strict schemas, least-privilege credentials, idempotency, and clear failure modes.

Data layer. Design PostgreSQL schemas that hold up under multi-tenant load and years of client data. Build ingestion pipelines for documents, structured data and enterprise sources; own chunking, embedding, indexing and the vector, graph and ontology structures that make retrieval accurate. Write the SQL that analytics agents generate and know when they are generating it badly.

Enterprise integration. Integrate with Microsoft Graph, SharePoint, Salesforce and client identity providers using OAuth 2.0 / OIDC, including delegated and On-Behalf-Of flows so agents act with the user’s permissions, not a service account’s. Respect the client’s existing RBAC and data boundaries in everything you build.

Multi-tenant product engineering. Build features that are tenant-isolated by construction: row-level security or schema-per-tenant where appropriate, tenant-scoped credentials and vector indexes, per-tenant cost attribution, audit logging that a client’s security team will accept.

Front-end. Build the interfaces through which users work with agents chat and task UIs, streaming responses, citations and provenance, review-and-approve flows, dashboards and admin consoles in React/Next.js and TypeScript, to a standard you would put in front of a client executive.

Quality and evaluation. Write the tests. Build eval sets for every LLM-driven feature and run them on every change. Instrument agents with tracing so that when a client asks “why did it say that”, you can answer.

What we are looking for

Must have

• 5-8 years building production web applications, with the last 2+ spent as a full-stack engineer owning features from database to UI.
• You have shipped LLM features to real users RAG, tool-calling, structured extraction or agent workflows using Claude, OpenAI, Bedrock or Azure OpenAI APIs, and you can talk concretely about what went wrong: hallucinations, latency, token cost, prompt regressions, retrieval misses, and how you measured and fixed them. • Strong Python backend engineering: FastAPI or similar async frameworks, typed code (Pydantic, mypy), backgrou…