The Normalizer

RFQs and MTOs come in every format imaginable: different languages, abbreviations, column layouts, units. Normalization means parsing each line, interpreting what product is being described, detecting errors (wrong schedules, impossible sizes, incompatible ratings), and matching it to your inventory. The engine doesn't just read text. It understands piping: it knows that SCH 30 doesn't exist in B36.10, that a Class 150 flange won't bolt to a PN 40 valve, and that "A106 Gr.B" and "P235GH" are not interchangeable. Both the RFQ and your inventory are normalized the same way, so matching is automatic: the engine finds what you have in stock, generates the quote, and flags what's missing.

Try it yourself: upload an RFQ and your inventory to ChatGPT and ask it to match items and generate a quote. You'll get a result that looks convincing but is full of wrong matches, invented specs, and missed items. LLMs are probabilistic, so every time you run the same request you get a different output. You'd have to review every single line manually, which defeats the purpose. Our engine is deterministic and rule-based: same input, same output, every time. Every match is traceable, every error is flagged, and the quote is generated automatically.

Real RFQs are full of abbreviations (SMLS, BW, RF, SW), typos, OCR artifacts, truncated descriptions, and inconsistent formatting. "A106 GR.B", "A106-B", "A106 Grade B", "ASTM A 106 GRB" are all the same material. The engine doesn't do simple string matching. It parses each description semantically, recognizes known abbreviation patterns, tolerates common misdigitations, and resolves ambiguities using piping industry rules. If something is genuinely unreadable, it flags it instead of guessing.

It flags the item for manual review instead of guessing. If a description is too ambiguous, has conflicting specs, or refers to a product outside the supported range, the engine marks it clearly and tells you why. No silent failures, no invented codes.

The engine uses dimensional tables from ASME, EN, and ISO standards to resolve sizes. A pipe described as "168.3mm OD x 7.11mm WT" and one described as "6 inch SCH40" are the same product. The normalizer maps both to a single code because it knows the actual wall thickness and OD values from standard tables, not from pattern matching.

A deterministic hash (SHA-256) generated from the normalized technical attributes of a product. Two descriptions that look completely different but describe the same pipe, fitting, or valve will always produce the same code. This is what makes automated matching possible: if the code matches, the product matches. Over 3 million unique codes are currently mapped.