Ujjwal Kanth
Personal notes, ramblings and occasional opinions.
Harness — Principal Engineer
Oct 2025 – Present
Still relatively new here, but already shipped something I’m pretty happy with.
Global Notification Service for Feature Flags
Built an SSE-based notification service that holds ~7 million concurrent connections for the Split.io SDK — delivering real-time feature flag updates with no SDK changes required. Replaced Ably as the vendor, cutting costs by over 60%.
The fun parts: thundering herd on mass reconnect, fan-out delivery guarantees at scale, zero-downtime rollout on a live production system. Stack: Go, NATS, ZeroMQ, inter-service communication over Unix Domain Sockets.
Next up: Agentic AI.
Unbxd / Netcore — 11 years, 8 months
Jan 2014 – Oct 2025
Unbxd is a Search & Discovery platform for e-commerce. I joined as a Software Engineer and left as Director of Engineering and Chief Architect — which in practice meant I wore every hat at least once.
Principal Architect (Jan 2022 – Oct 2025)
Search Index Improvements Improved query performance by 20–30% and cut costs by ~10% by overhauling the search index infrastructure. Introduced a new Query Engine with a custom DSL, custom replica placement and replication, and upgraded the stack to ARM + Apache Solr 9.
Data Pipeline v2 Next-gen data pipeline built to address scaling problems the old one simply couldn’t handle anymore. Apache Flink for streaming, Apache Druid + Superset for warehousing, CockroachDB for reporting. Custom Flink operator for dynamic on-demand scaling brought operational costs down 30%.
Catalog Ingestion Pipeline v3 A highly scalable, event-driven pipeline handling everything from a few thousand to hundreds of millions of catalog products. Built on Argo Workflows on Kubernetes with Spot Instances — cost reduction of 40–60% depending on load.
Document Store High-speed catalog storage for query-time retrieval. Cut serving latency by 20% by trimming data overhead in the stack. Built on PostgreSQL + Aerospike.
Product Features User-level segmentation and merchandising services; traffic classification using Redis + CockroachDB; schema and lifecycle management improvements; analytics reporting on BigQuery.
Architect (Jan 2019 – Jan 2022)
Inter-Region, Inter-Datacenter Network Topology Designed and built the inter-region communication layer for all microservices — event-driven, using NATS for configuration, metadata, and event propagation. Routing is based on index affinity across regions and datacenters.
AI Data Pipeline Custom data lake on S3 + Parquet for training ML models. Kubeflow connectors for easy pipeline access.
Standardisation of Internal Libraries Open-source Go toolkit: unbxd/go-base. Internal Java starter project and utility library on Quarkus.
Senior Software Engineer (Jan 2014 – Jan 2019)
Edge Server — Custom API gateway handling all network operations, observability, and logging.
Centralised Configuration Management — Hierarchical rule system across tiers and groups, built on DGraph + NATS.
Query Pipeline — Migrated a monolith to microservices. Custom Solr 6.6 fork, custom query parser, similarity class, and search components.
Other notables — GitOps-based CD with Flux CD; internal API gateway for inter-team calls; Protobuf + gRPC adoption for inter-service communication (20% latency improvement); contributions across Orchestrators, Query Analysis, Promotions, Analytics Pipeline v1, and more.
Before Unbxd
| Company | Role | Period |
|---|---|---|
| Parablu Inc. | Product Engineer | May 2013 – Jan 2014 |
| Manhattan Associates | Software Engineer | Jul 2012 – May 2013 |
| Tata Consultancy Services | Asst. Software Engineer | Jul 2010 – Jul 2012 |
Full details in the resume.