Dmytro Galko · open-source tools

llm-doc-extractor

Name: llm-doc-extractor
Author: Dmytro Galko

Stop retyping fields out of invoices by hand

Every company does the same loop: open a PDF or an email, find a handful of fields, type them into something structured. It is the most widely duplicated manual work there is. This does that loop once, repeatably: you describe the fields in YAML, and the tool returns exactly those keys every time - across Anthropic, OpenAI or a local Ollama model with one flag.

Source on GitHub Work with me

What it does

A built-in demo provider runs the whole pipeline with no API key, so the shape can be tested before a vendor is chosen.
The interesting part is the failure: a reply that does not fit the schema is caught, reported and retried against the declared shape, never written downstream.
One schema, three providers - so a price rise or an outage at one vendor is a flag change, not a rewrite.

How it is proved

The sample invoice returns invoice_number, date, vendor, total and line_items as typed JSON with no key missing and no key invented. Same schema, same output shape whether the call went to Anthropic, OpenAI, a local model, or the offline demo provider.

PythonYAML schemaAnthropic / OpenAI / Ollamaruns offline with no key

People find this looking for

extract fields from invoices with an LLM, document to JSON schema, provider-agnostic LLM extraction, Anthropic OpenAI Ollama one schema, test document extraction without an API key.

If that is your week and you would rather someone else did it, write to hello@dkautomation.dev or open an issue.