A portfolio should read fast and still reward curiosity.
Self-hosted OpenAI-compatible gateway for org-wide LLM calls: routing, governance, cost control, and observability.
Problem: Teams scatter provider keys across apps with no shared auth, budgets, fallbacks, or request visibility.
Solution: Built a Fastify data plane with API-key auth, multi-provider adapters, rule-based routing, caching, rate limits, and basic guardrails, plus a Next.js admin dashboard.
Impact: Gives organizations a single control plane for LLM traffic without locking into a managed proxy.
My role: Designed the monorepo architecture and implemented the gateway hot path, workers scaffolding, and dashboard.