Results-driven Engineering Manager known for high productivity and efficient task completion. Specialize in project management, process optimization, and team leadership. Excel in communication, problem-solving, and strategic planning to deliver exceptional results in complex engineering environments.
Backend Engineering, Distributed Systems, Search & Recommendation.
Leading a team of engineers to build and scale a highly available, low-latency search platform handling 170M+ requests/day (P99: 350ms).
Focused on mentoring, coaching, and fostering a collaborative culture that emphasizes collective success.
1. Rebuilt the re-ranking platform into a configurable system supporting 40+ ranking models without requiring code changes.
2. Developed a scalable query understanding layer supporting 100+ attributes across product types, driving relevance across queries.
3. Driving operational efficiency: building a Search Agentic Tool to debug issues and help product/ops teams understand search behavior. Redefined test strategies to improve release quality and developer confidence.
4. Re-architected candidate retrieval to enable live vector search with dynamic switching between vector and lexical strategies based on runtime signals, optimizing for latency and cost.
5. Redesigned pagination flow with a unified system that dynamically fetches lexical, vector, and explore candidates to generate category-aware pages, supporting deep pagination.
6. Re-architecting ranking sublayer to optimize multiple business goals using a no-code, rule-based configuration setup.
7. Building an experience versioning and failure-trigger system that leverages DS signals to validate relevance, with automated rollback to stable versions.
8. Ensuring high availability via initiatives like index corruption handling, multi-AZ fault tolerance, and flow-specific rate limiting and isolation.
9. Led the initiative to introduce ScyllaDB at Zepto and implemented the first use case for search ranking. The system ingests multi-level signals (store-product, city-product, and store-level) to power ranking algorithms with sub-millisecond latency. Successfully implemented a high-performance solution supporting 500K+ requests per minute with exceptional read/write throughput.
English