Architecture, code, infrastructure, and operations — built and run solo, on my own infrastructure. A live AI SaaS with paying subscribers and a ~40-service autonomous agent fleet, on top of a corporate finance career.
A production AI SaaS built and operated solo. Multi-model LLM pipeline (Anthropic Claude + open-weight models) with retrieval-augmented generation, structured outputs, and deterministic validators that catch hallucinations before users ever see them.
The hard part isn't generating text — it's gating it. Every paid reading passes claim-level validation against computed source data before it ships.
Next.js/React/TypeScript front end, Python/FastAPI services, PostgreSQL + Redis, Stripe subscriptions with multi-currency pricing and 13-language i18n — fully self-hosted on Linux.
~40 always-on services across Linux VPSs on a Tailscale mesh: autonomous Claude-based agents running real operations, data, and content workflows — with error handling, budget controls, health checks, and alerting, because unattended agents without guardrails are a liability.
Custom MCP servers, RAG over PostgreSQL/pgvector, and a LiteLLM multi-model routing gateway (Anthropic + NVIDIA NIM + local Ollama) with hard per-model spend caps.
Healthcare-adjacent SaaS: a dental-practice intelligence platform. Node/Prisma on PostgreSQL (70+ table schema), role-based access control, audit logging, Stripe billing — built end-to-end.