ZOMATO AI ANALYTICSZOMATO AI ANALYTICS

The data starts as 7 spreadsheets (orders, menus, restaurants, reviews, and more) sitting in Amazon S3 — about 33.6 million rows in total. From there it flows into Snowflake, a cloud data warehouse, in stages: first loaded exactly as-is, then cleaned and given proper types, then finally shaped into simple, ready-to-use tables that answer real business questions like "which restaurants make the most money" or "how often do deliveries run late." Every step is checked automatically — 17 models and 16 automated tests make sure numbers stay correct as the data changes.

On top of the clean data sit three AI features powered by Google's Gemini models. First, it reads through customer reviews and automatically tags each one with sentiment and complaint type — smartly, since only 294 of the 300,000 reviews are actually unique, so it does the AI work once per unique review instead of 300,000 times, cutting the cost by over 1,000×. Second, it can search reviews by meaning, not just keywords. Third, it lets someone type a plain-English question and turns it into a safe, read-only database query — with checks in place so it can never accidentally change or delete data.

Security is treated as seriously as it would be at a real company: cloud storage access uses temporary, rotating credentials instead of permanent passwords, and every pull request automatically runs six checks — code-quality scans, a check that no secret keys were accidentally committed, and a check that the sample data used for testing still lines up correctly — before anything can be merged.