At Flipkart, I improved regression coverage for Shopsy’s Semantic Search pipeline with a layered test suite covering more than 2,000 query scenarios. I also implemented automated layer-wise flagging that reduced root-cause analysis time by over 50%.
On TickLadder, a team project, I built a single-threaded C++20 limit order book matching engine that reached 51M messages per second, with p99 latency of 81 ns and zero hot-path heap allocations. I verified it with 83 tests, including differential and mutation testing.
For the Amazon ML Challenge 2026, I worked with Team Neural Ninjas on an entity-resolution pipeline matching business records across three sources and three countries. I also engineered a ScaNN retrieval mode that retained 99.9%+ of matches at 0.3–1.9 ms per query.
