Bathinda, Punjab, India
I build things at the layer most people don't see — indexes, compilers, shells — and I like it there. At Teradata, I built a compute-local indexing feature for the query optimizer that delivered a 2.7× throughput improvement on a 2 TB benchmark, cutting object-store round-trips from three to one. That meant working across cost modeling, execution plans, and DML maintenance under real production constraints. On my own time: a Unix shell in C, a Standard ML compiler emitting three-address code, and an independent extension of a graph-reachability index to handle probabilistic edges via Monte Carlo sampling. 300+ LeetCode problems solved along the way. B.Tech CS, IIT Jammu. Currently looking for SDE-1 roles — systems, databases, or backend. Let's talk.
Query optimizer team, working on performance and correctness for Teradata's MPP database engine on public-cloud deployments. Designed and shipped a compute-local indexing feature that delivered a 2.7× throughput improvement on tactical (low-latency, point-lookup) queries on a 2 TB benchmark, by eliminating cluster-wide metadata traversal of cloud object storage and cutting object-store round-trips from three to one per query. Built the index to materialize object-location metadata on fast local storage keyed by a stable logical identifier, replacing a multi-step, all-node query plan with a bounded few-node access path for cloud-object-store-backed tables. Implemented end-to-end index-maintenance workflows in C++ for INSERT, UPDATE, and DELETE, handling immutable-object rewrites and the distinction between physical (row-table) and logical (columnar-table) deletes. Architected the maintenance logic as a single extensible module using a composable design pattern, so new scenarios like storage reorganization could reuse existing building blocks. Drove changes across the query optimizer and execution engine to integrate the new index as a fast access path, including cost modeling and plan-generation changes spanning parsing, resolution, optimization, and execution. Diagnosed and fixed a subtle correctness defect in delete-triggered index maintenance using core dump analysis and optimizer-level tracing, then built out unit, functional, and regression tests to lock in the fix with zero impact on existing query behavior.