What it does and how it works.
RAGDoll is a simple way to get answers from your own files. Upload anything — documents, spreadsheets, slides, even scanned images — and just ask, in plain words. Every answer comes with the exact page it came from, so you can always check it's right.
Documents, spreadsheets, slides, images and web pages — all searchable, with answers pointed back to the source.
Your files are yours alone. Nothing is shared with anyone, and you can export or delete everything anytime.
Ask a question and get a clear, direct answer fast — no scrolling through long files to find it yourself.
A portfolio project showing real, end-to-end engineering. View on GitHub →
Uploads are processed in the background by Celery workers with a Redis queue, so the app stays fast while big files are parsed, chunked and embedded.
Sign-in uses Supabase (ES256 JWTs verified on every request). Each user only ever sees their own files and conversations.
A ReAct agent can search your documents, search the live web (DuckDuckGo) and do maths — choosing tools itself to answer harder questions.
Every request is traced with OpenTelemetry to Langfuse Cloud, with Prometheus metrics and structured logs — so nothing is a black box.
Redis-backed rate limiting, health checks, and disconnect-safe streaming keep the service stable under real traffic.
GitHub Actions runs lint, type-checks and tests on every push. Shipped as Docker containers behind Caddy (automatic HTTPS) on Oracle Cloud.