Baresente

Outcome

A ~35-endpoint API across nine routers, with user-scoped data, receipt OCR, and cached LLM spending insights that fall back deterministically when the model is unavailable.

What I built

• Designed the application around 10 relational models with user-scoped constraints, composite indexes, and versioned Alembic migrations.
• Built a ~35-endpoint REST API across nine routers with JWT authentication, dashboard aggregation, and receipt OCR.
• Cached LLM spending insights with deterministic fallback handling so the dashboard still works when the model is unavailable.
• Containerised the service and wrote deployment configuration for both Render and Fly.io, with PostgreSQL as the system of record.
• Set up CI/CD workflows using GitHub Actions and Pytest to support reliable releases.

Stack