Architecture Overview

Back to index · Related: vision, rfq-anatomy

High-level shape of the system. This is a starting sketch — nothing is committed yet.

Pipeline

RFQ source ──▶ Ingest ──▶ Extract ──▶ Analyze ──▶ Assist quoting
 (PDF/email/    (parse,     (map to     (complete-    (line items,
  doc/text)     normalize)  schema)     ness, risk)   handoff)

Ingest — accept and normalize input formats (start with PDF).
Extract — pull structured fields per rfq-anatomy.
Analyze — completeness, ambiguity, risk, feasibility.
Assist — surface comparable line items and export toward a quote.

Open architectural questions

Document parsing approach (OCR? layout-aware? LLM-based extraction?)
Where does human review fit in the loop?
Data model for requirements and line items
Multi-tenancy and per-customer data isolation (it’s a SaaS)

Non-functional considerations

Privacy — RFQs may contain commercially sensitive data.
Auditability — extracted data should be traceable to source.
Extensibility — events first, but the model should generalize.