Node.js/Claude API Developer — Security, Reliability & Cost Optimization for AI Research Tool

We run a live, working product that sources and verifies companies against custom criteria using an LLM-powered research pipeline. It was built solo by a non-technical founder and works — real research runs happen on it today — but it needs a developer to hardn it, fix real known issues, and bring the cost and reliability up to a standard we can put in front of paying clients. Stack: Vite + React 18 frontend, Node.js/Express backend on a VPS behind nginx, Supabase/Postgres for data, Claude (Anthropic) as the sole AI provider — including a multi-round tool-use loop that does the actual research and verification work. What needs doing, roughly in priority order: Security — review how credentials and API keys are handled, and specifically review a service-role database access pattern where ownership is enforced in application code rather than at the database level Reliability — there's currently no crash-safety net (no global error handlers, no health-check endpoint) and no automated tests or CI at all Cost optimization — the AI pipeline currently costs roughly $0.20–0.35 per unit of work processed, and one step alone averages tens of thousands of input tokens per call in a tool-use loop. We need someone who can actually dig into why and cut it, not just add caching in general General quality — broader improvements to research accuracy, output quality, and the dashboard UI/performance, as time and budget allow This is an audit-and-fix engagement, not a rebuild. A full written technical brief and system audit already exist — happy to share both with serious applicants before you quote, so you're scoping against real information, not guesswork. How we work: Async, no fixed hours. Check in before touching anything production-facing; otherwise work at your own pace and document what you change.