Work · Client project
AI product-catalog enrichment
From a barcode and a name to a full product record: Arabic and English descriptions, SEO fields, and matching against supplier lists.
The challenge
Stores had hundreds of products with no description, no images and no SEO fields, just a barcode and a name. I built tools for several clients that fill in this data automatically.
What I built
- Full product record: from the barcode and Arabic name to Arabic and English descriptions and length-limited SEO fields, in one request to Gemini.
- Quality guard: rejects descriptions that contain invented claims, retries, and resumes where it stopped.
- Supplier matching: barcode first, then fuzzy matching in three tiers, then Gemini checks each pair.
- Images: finds image links for products.
- Cost calculator: estimates the running cost before a job starts.
Tech
Command-line tools in Python, with Excel files in and out. Gemini API for writing and checking, pandas for data handling, and RapidFuzz for fuzzy matching.
Outcome
I built these tools for several clients between August 2025 and February 2026. In one job, 482 of 488 products were completed; in a matching job, 1,321 of 1,363 products were confirmed. Client names are withheld, and the images use a product I made up.
- Type
- Client project
- Services
- Connecting your systems
- Tech
- Python
- Gemini API
- pandas
- RapidFuzz
Screens
Tell me about one process
If a process in your company still runs on spreadsheets and group chats, tell me about it. I'll tell you honestly whether a custom system is worth it.