A search that understands “pet dard”.
- My part
- Problem, spec, rules, testing
- Code
- AI coding tools
- Tests
- 50,000+ searches
The problem
We paid about $300 a month for a search app on our India and USA stores. A customer who typed “pet dard”, Hinglish for stomach ache, got nothing useful, because no product page uses those words. Ayurveda products also help with more than one thing, so one product should show for several needs.
I could have kept the paid app. It was hurting the customer experience, so I set out to replace it.
What it has to handle
| Example search | What it must do |
|---|---|
| pet dard | Understand Hinglish and find the right products. |
| sleep, stress | Show the right product, with a line on what it is for. |
| chest pain | Show emergency numbers, 112 and 108. No products. |
| anxiety | Show the Tele-MANAS helpline, 14416, and a doctor link. |
What I decided
- It reads English, Hinglish, Hindi and Malayalam.
- It only shows products from our own catalogue.
- Health searches follow fixed rules, as in the table.
- The brand team can pin, move or hide products without code.
How I tested it
50,000+ test searches, mixing real searches from our search history with ones I wrote. It passes all of them today. I wrote some of those tests myself, so real shoppers are the test that counts.
I also built a scoring sheet to compare it with the paid app on the same searches. It returns the right products for each search, and more relevant ones than the paid app. It is in testing on our US store. India is next.
How it was built
I wrote the spec and the rules and directed AI coding tools. I did not write the code.