AI RFQ processing, with a deliberate stop.
A synthetic experiment in turning unstructured commercial requests into governed record drafts—without confusing confidence with authority.
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
- 01Incoming email
- 02Classify
- 03Extract
- 04Match entities
- 05Draft records
- 06Human review
- 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
Authority matrix, tool contracts, review card and evaluation for this agent