Lab / 001 · In progress

AI RFQ processing, with a deliberate stop.

A synthetic experiment in turning unstructured commercial requests into governed record drafts—without confusing confidence with authority.

Synthetic lab

Independently created. Contains no employer or client implementation detail, internal names or figures.

01Hypothesis

AI should assemble the decision surface—not make every decision.

Classification and extraction can remove clerical effort. Customer identity, product ambiguity and commercial commitment remain reviewed, because their failure consequences cross system and human boundaries.

02Workflow

  1. 01Incoming email
  2. 02Classify
  3. 03Extract
  4. 04Match entities
  5. 05Draft records
  6. 06Human review
  7. 07Commit

03Control model

Automate

Reversible work

Classification, field extraction, candidate retrieval and draft assembly.

Review

Ambiguous work

Entity resolution, product substitutions and low-confidence quantities.

Authorize

Consequential work

Pricing exceptions, commitments and writes to operational systems.

04Current status

  • Synthetic email corpus and evaluation cases defined
  • Structured extraction contract prototyped
  • Candidate matching with explanation in progress
  • CRM commit intentionally disabled until review controls pass

05The design behind it

AI & Agentic Workflows · flagship caseAuthority matrix, tool contracts, review card and evaluation for this agent