Full-Stack AI Engineer (Node.js/TypeScript) — Integrate AI Agents into Existing SaaS Platform

We run a B2B SaaS platform (Node.js/TypeScript backend, React frontend, PostgreSQL) with an established customer base. We're adding an AI layer to it and we need an engineer who can do both halves of that job properly. THE PROBLEM Our users spend hours on manual work inside our app — searching records, summarizing documents, moving data between our system and their tools. We want AI to absorb that work. We do not want a chatbot bolted onto the corner of the dashboard. We tried this once already with an "AI developer" who built a great demo that fell apart the moment real users touched it. No error handling, no cost controls, no tests. We're not repeating that. WHAT YOU'LL BUILD - AI features integrated directly into our existing codebase — natural-language search over customer data, document summarization and extraction, automated classification - An agent layer that can take action through our existing API, not just answer questions - A RAG pipeline over our customers' document sets, with citations back to source - An MCP server exposing parts of our platform so customers can reach it from Claude and ChatGPT - Cost tracking, rate limiting, and fallback behavior so this is predictable in production OUR STACK Node.js, TypeScript, NestJS, Express, PostgreSQL, Supabase, React, Docker, AWS. OpenAI primary, evaluating Anthropic. WHO WE'RE LOOKING FOR - 8+ years in Node.js and TypeScript — you've maintained production systems, not just launched them - Real experience with OpenAI API, LangChain.js, and vector databases (Qdrant, pgvector, or similar) - You design the database schema and architecture before you write the feature - You write tests. Non-negotiable - Strong written English and the ability to push back when we ask for something dumb - Top Rated or 90%+ Job Success preferred NICE TO HAVE - MCP server development - Playwright or browser automation experience - Prior work turning a prototype into a production MVP HOW WE WORK Weekly milestones, work in our repo on branches, daily pushes, async-first with one call a week. We expect to see progress every week — no six-week black boxes. This starts as a 3-month engagement at 30+ hours/week with a strong chance of continuing. We're looking for someone to own this layer long term. TO APPLY Skip the template. Tell us about one AI feature you shipped into an existing production application: what it did, what broke, and how you handled it. If you've built an MCP server, lead with that.