Finance

Statements, invoices and filings at volume, where cost per page and a reviewable trail both matter.

A lens over 4 jobs. Draft copy, under review; the models and numbers on each job page come from today's data.

The jobs it bundles

Each links to its job page (or the nearest scoreboard view while the page doesn't exist yet).

Pull fields from thousands of statements at the lowest cost per page.

102 models fit

Scanned invoices and tables into JSON.

74 models fit

Read a full annual report or prospectus in one pass.

238 models fit

Tag transactions and route exceptions.

66 models fit

Each constraint opens a job page or the scoreboard with the filter already in the URL.

Cost at volume

At millions of pages, the price per 1,000 pages decides more than the last point of accuracy.

Hosted, cheapest per 1,000 pages →
Data residency

Documents can't leave your own hardware or region, so start from models you can run yourself.

Self-host only → · Fits one 24 GB GPU →
Audit trail

Prefer models whose behaviour we checked on a live endpoint, with the probe, the reply and the date on file.

Models with first-look receipts →

How people wire a model into this job. Notes are written by hand from the cited cards and docs; the model links go to the facts.

Model: BAAI/bge-m3
Harness: Embedding endpoint (a serverless provider or self-hosted TEI) + any OpenAI-compatible chat endpoint
Weights: FP16 embeddings; FP8 generator

Chunk to the embedding model's max sequence length (card says 8192 tokens for bge-m3).
Use the generator's structured-output flag (provider row) for JSON extraction.
Re-embed on model change; keep the embedding model id in the index metadata.