The problem
A data team structured and checked very large numbers of product listings by hand. Researching a missing brand path, filling brand, size and category fields, and comparing a merchant's record with the standard record were all slow and easy to get wrong.
What I built
Three tools, used by the team:
- Brand lookup: a search over one central brand reference list, with type-ahead. It comes as a web page, a Chrome extension and a desktop app, and it replaced researching each missing brand path by hand.
- AI-assisted categorization: turns raw listings into structured fields in batches, using AI plus brand and category rules, with a person reviewing unclear rows.
- Product database verification: paste rows of merchant and database product names and get a Match or Mismatch verdict, field by field, with reasons.
A related batch experiment used an AI model to sort a large brand master list into four-level category paths. Its output fed the brand lookup's category paths.
How it works
- Raw listings
- Rules and brand lookup
- AI for the rest
- Person reviews unclear rows
- Structured fields
What it changed
- Brand lookup: about 50 teammates used it, with over 80,000 lookups logged in about five weeks (Aug to early Sep 2025, web version). I estimate it saved 1 to 3 minutes per missing-path lookup (my estimate).
- Categorization: batches of about 50 products went from roughly 1 to 1.5 hours to usually under 45 minutes (my estimate).
- Verification: a typical 100-line review went from about 15 minutes to 2 to 3 minutes (my estimate).
My role
I built and introduced the tools, defined the workflow and rules, and used AI-assisted development for the code.
Skills shown
Flask, PostgreSQL and SQL enrichment, AI prompting and batch processing, rule design, Chrome extension, internal tooling adopted by a team, working with AI tools.