the problem
Teams still retype invoices, contracts and forms by hand. Each supplier uses a different layout, so fixed templates break as soon as a document changes.
what it does
halyard reads a document, finds the fields you care about and returns them as structured data. Anything it is unsure about goes to a person for review, with the source text highlighted.
how it’s built
A Python service runs extraction in stages: layout, fields, then validation against your rules. Every result keeps a link to where it came from, so each value can be checked.
current status
Live, with extraction accuracy tracked on every release.
tech stack
- python
- fastapi
- postgres
- llm extraction