How a Catering Parts Desk Answers 90% of Part Searches

Salespeople handled thousands of quote requests each month, but technical drawings and parts lists were spread across an old shared drive. Finding the right part involved manual document searches or waiting for a supplier, delaying replies to customers. The distributor needed a shared way to find parts and check the source drawing before quoting.

Stackfix built a shared AI search over the distributor's manufacturer partbooks and technical drawings, running alongside the existing ERP. Gemini Flash reads each partbook PDF and extracts the parts, diagrams and machine models into an index. Search runs on Postgres text and fuzzy matching over that index, with GPT-5.4 interpreting what the salesperson typed and ranking the candidate parts. When a part isn't in the catalogue, Perplexity Sonar looks it up on the web. Every result links back to the original drawing, so the salesperson checks compatibility before the part goes on a quote. Newly uploaded drawings become searchable for the whole team. The system is model-agnostic, and Stackfix swaps in better models as they arrive.

Stackfix started with the distributor's archive of manufacturer partbooks and technical drawings, which sat on an old shared drive. Gemini Flash was set to read each PDF and pull out the parts, their diagrams and the machine models they fit, building an index of more than 1.6 million parts. Search was built on Postgres text and fuzzy matching over that index, with GPT-5.4 interpreting each query and ranking the answers, and Perplexity Sonar added for web lookups when a part isn't in the catalogue. Every result links to its source drawing so the salesperson verifies the match before quoting. The system went live alongside the ERP in four to eight weeks, and the team now rates searches so accuracy can be tracked.

Distributors, manufacturers and aftermarket service teams that handle frequent parts enquiries using large collections of technical drawings, manuals and parts books. It is particularly relevant where salespeople spend time identifying components across many suppliers or equipment models, and can verify a proposed match against the original documentation before quoting.