A market priced in chat messages

Problem

Iran’s phone and accessories wholesale trade is priced through dozens of dealer chat channels. Every dealer names products differently, and prices swing with the dollar. Sourcing meant a person cross-checking channels by hand.

What I built

Price lists arrive as text, PDF and images. Each line is read, matched to one product catalogue and one dealer, and turned into the best source per product, price trends, arbitrage signals and a trust score per dealer. Matching works across Persian, English and transliterated names.

Measured

• 22,665 canonical products
• 3,421 dealers
• About 43,000 dealer-list observations spanning 34 months
• Exchange-rate history back to 2011
• Page response cut from 6.35s to 0.49s; worst case from 43.8s to 0.85s