RateParse.

AI that reads freight rate sheets and answers "what's my rate for this lane"?

RateParse uses AI (a Claude-based extraction model) to read carrier rate sheets during upload/parsing, but lane-rate lookups themselves run as fast database queries — not a live AI call — so answering "what's my rate for this lane" returns in under 2 seconds without per-query AI cost.

Two different AI roles in this workflow

Why this matters for cost and speed

A design that calls an AI model on every lane search would be both slower (AI inference latency) and more expensive per query than a structured database lookup. Splitting the AI work to parse-time only means search stays fast and the per-sheet AI cost is bounded and predictable.

How RateParse implements this

RateParse caps parse cost at $0.50 per sheet via a token budget and a single regeneration retry; once a sheet is parsed, lane search runs as a Firestore query with your margin rule applied, returning results in under 2 seconds at p95 with no AI call on the search path.

Frequently asked questions

Does every lane search cost money in AI usage?

No — AI/LLM usage happens only when a rate sheet is parsed (once per upload). Lane searches query the already-structured database and do not invoke the AI model.

How accurate is AI extraction of rate sheet data?

Extraction output is validated against a fixed schema before it is stored; if validation fails twice, the sheet is flagged with a per-row error summary instead of silently storing an incorrect or partial rate.