I work on Zomato's Infra team — cloud architecture across every service, AWS, and the shared Go libraries (logging, config, gRPC, caching, messaging) that other teams build on without thinking about.
I like taking a systems paper and turning it into working, benchmarked Go code — that's where most of my open-source time goes.
RibbonFilter/ribbonGo · Creator & Maintainer
First pure Go implementation of the Ribbon filter (Dillinger & Walzer, 2021). 4.7% overhead above the information-theoretic minimum, 12.7% more space-efficient than Bloom filters at equivalent FPR. SoA memory layout, software-pipelined prefetching, zero heap allocations on build and query hot paths.
RoaringBitmap/roaring · Contributor (2.9K+ stars, used by InfluxDB, Bleve, DataDog)
- feature:
CardinalityInRange- binary search based range counting, up to 99% faster than full-scan Rank approach on narrow ranges - perf: In-place XOR - across array, bitmap, and run containers; 44% speedup, 100% fewer allocations
- perf: Galloping intersects2by2 - skip-ahead for skewed set sizes (64x ratio)
valkey-io/valkey-glide · Contributor (Go client)
- perf: Dropped FFI call from response type checks - perf optimization in the Go client hot path
- feature: AzAffinityAllNodes read strategy
valkey-io/valkey-go · Contributor (fast Valkey/Redis client with auto-pipelining and client-side caching)
- bug: Re-pick unfinished commands between retry iterations - fixed an unbounded retry loop in
DoMulti/DoMultiCachewhen a replica drops mid-flight; rebuckets against the refreshed slot map, preserves ASK routing, keeps transaction spans coherent. Regression tests for replica removal, mixed ASK/transport errors, and response ordering. - perf: Consolidate keyless commands onto one conn on retry - stops fanning out keyless commands across arbitrary picks during retries
- feature: Opt-in
PreferClusterShards- useCLUSTER SHARDSahead of version 8 on patched servers, avoids routing to nodes still loading during rolling upgrades