You can upload invoices, have them parsed, and correct the parsed data. The corrections are saved, and the next time you upload a similar invoice, the agent may apply them automatically if they make sense.
You can provide feedback on the automatically applied corrections. Rejected corrections are immediately reverted, and you can add an explanation, which will be taken into account the next time the agent encounters a similar invoice.
Here's a 4 min demo showing the full feedback loop in action: https://www.youtube.com/watch?v=txaJ0OhOFVw
Export as JSON or CSV. You can also use it via the API.
Tech stack: Postgres (with pgvector), FastAPI, Claude for extraction, Celery + Redis for background processing, TypeScript with React, Voyage AI for embeddings, Mistral for OCR (with pymupdf fallback).
It's pay-as-you-go, no subscription required, just top up your account and use as needed. You can try it for free.
I'd really appreciate hearing what you think, especially if you deal with invoices regularly. Is the correction/feedback loop something that would actually save you time?
Thanks! Martim