RFQflow

Inspiration

Working with different customers and manufacturers showed us how easily quotation details can get lost or misunderstood. Buyers describe materials in different ways, and manufacturers’ quotations do not always match the original request. These mismatches can delay orders and complicate payment discussions. We built RFQFlow to make the path from buyer request to quotation clearer and more reliable.

What it does

RFQFlow turns messy RFQs from text, PDFs, spreadsheets, and emails into structured material specifications. It finds supplier catalog candidates, checks compatibility, calculates demo pricing, and flags missing compliance evidence.

When specifications are ambiguous, it asks the buyer targeted questions instead of guessing. After the buyer answers, it resumes the same request and generates a quotation only when the implemented readiness checks pass.

How we built it

We built a Python backend with FastAPI and a Streamlit interface. Nicrron powers specification extraction and embeddings, while ChromaDB retrieves supplier candidates. Pydantic validates structured data, and deterministic rules handle compatibility, compliance blockers, and pricing with Decimal arithmetic.

Our prototype uses a clearly labeled synthetic catalog and supplier costs with a documented 20% markup. It does not claim live market competitiveness.

Challenges we ran into

Industrial descriptions contain abbreviations, gauge thicknesses, fractions, and dimensions without units. Interpreting “16ga 4x8” safely required preserving uncertainty rather than assuming a standard size.

We also had to connect document extraction, retrieval, and quotation processing while preserving source references. Another challenge was separating a semantically similar product from a compatible product—and a requested certificate from verified supplier evidence.

Accomplishments that we're proud of

We built and demonstrated the complete workflow: RFQ input, extraction, supplier matching, clarification, pricing, and quotation generation.

The buyer can resolve ambiguous specifications without losing the original request or starting a new session. Missing compliance evidence blocks quotation readiness, and our automated suite passes 27 tests.

What we learned

Reliable AI applications need more than a convincing model response. Language models help interpret messy requests, but business decisions need clear rules, traceable evidence, and reproducible calculations.

We learned that asking a precise clarification question can be more valuable than producing an immediate answer.

What's next for RFQflow

We plan to connect real supplier catalogs, current supplier prices, and comparable market benchmarks. We also want to add verified certification records, persistent quotation history, and stronger extraction evaluation.

Further work includes scanned-document OCR, broader product support, and production access controls. Our priority is turning the demonstrated workflow into a dependable tool for real industrial sourcing