Manufacturing · Document AI · 2025

TriCab RFQ parser — unstructured quotes, made structured

A production AI system inside a cable manufacturer that turns unstructured request-for-quote input into structured, quotable data. Client work — shown as workflow and architecture only, no company data.

By the numbers

Unstructured→structured
RFQ text to quotable fields
In production
running inside the business
Human approval
kept on every quote

The problem

Requests for quote arrive as free text — emails, PDFs, notes — every one formatted a little differently. Someone has to read each one and translate it into the specific fields the quoting process needs. It’s slow, repetitive, and easy to get subtly wrong.

The approach

I built an extraction system that reads the unstructured RFQ input and produces structured, validated fields ready for quoting — with a person still approving the result before it becomes a quote.

  • Extraction — an LLM-backed pipeline pulls the relevant fields out of messy input.
  • Validation — the output is checked against a schema and the business’s real constraints, so bad extractions are caught, not quoted.
  • Human-in-the-loop — the system drafts; a person confirms. The point was never to remove the estimator, only the re-typing.

The interesting engineering isn’t “call a model.” It’s making the output trustworthy enough that a business will put it in front of a customer.

The result

The parser runs inside TriCab today, turning input that used to be hand-processed into structured data the quoting workflow can use — with the approval step intact.

NDA boundary. This is live client work. I can walk through the architecture and approach, but no TriCab data, volumes, or customer details appear here — examples are synthetic. Happy to go deeper in a conversation, within what I'm allowed to share.