Data-ingestion platform
Music royalties business · Platform engineer · team of three
A full-stack data-ingestion platform (Django + React) for a music royalties business, now live with the engineering team. I moved a volatile codebase onto a Turborepo monorepo, migrated the frontend off an outdated CRACO build to Vite, and built its CI/CD from scratch: GitHub Actions with self-hosted runners, automated tests, and auto-deployment to AWS EC2, plus a dedicated EC2 test environment. I also prototyped a layout-fingerprinting technique (perceptual-hash style signatures) to identify statement sources where keyword matching fell apart.
DjangoReactTurborepoGitHub ActionsAWS EC2CI/CD
Collaborative desktop app
Music royalties business · Sole designer & developer
A native desktop app (Tauri/Rust + React) in daily use across a team to track incoming client data, replacing brittle spreadsheet and Monday.com workflows. Built as a realtime collaborative workspace on Supabase with a keyboard-driven interface, and owned end to end: from the UX through to a signed, auto-updating on-prem release pipeline. It also exposes a custom MCP server so a local LLM can scan raw data, extract metadata, and update the app directly.
TauriReactTypeScriptSupabaseshadcn/uiMCP
AI reconciliation agent
Music royalties business · Sole designer & developer
An LLM-driven agent that reconciles ingested royalty data against source statement PDFs and captures each statement's full financial movement. I re-architected a brittle regex approach into an autonomous agent with document retrieval and vision-based extraction, and cut running costs by routing each step to the cheapest capable model across OpenRouter. Correctness on non-deterministic output is held with an extensive automated test and evaluation suite.
PythonLangGraphLangSmithOpenRouterAWS S3