Apr 2023 – Jul 2025 · 2 yrs 3 mos · Bengaluru, Karnataka, India · Hybrid

• Led architecture and MVP delivery of Open Engines, enabling customers to spin up and run open-source compute engines (Trino, Flink, Ray). Designed backend infrastructure, drove cross-functional execution
• Owned and scaled Lakeview, the Hudi table observability system. Built Kafka/SQS-based streaming ingestion; matured system for multi-cloud, production-grade usage.
• Contributed to the design of Onehouse SQL and Spark support, including the metadata catalog and job infrastructure. Helped enable integration of Spark, PySpark, and SQL workflows into the Onehouse Console.
• Built backend features incl. adding support for MySQL CDC ingestion, secrets migration (firebase to secrets manager), and async job orchestration using in-house workflow engine (co-designed and stabilized)
• On-call contributor since early stage, led postmortems, improved monitoring and system stability.

Jan 2023 – Apr 2023 · 3 mos

Jul 2021 – Feb 2023 · 1 yr 7 mos

• Data-platform team
• A self service platform to configure table ingestion into our delta lake, create transformations and query/analyse/export the final data.The Platform was widely used by data/product analysts, data scientists, engineers, etc at a massive scale across Udaan
• Own and Maintain the End to End Table Ingestion Pipeline (UI + Backend): Rearchitected the entire ingestion logic and migrated to a custom in-house compute back-end framework leading to an approx 30% ingestion cost reduction
• Built and Maintained the DAG based execution Feature (UI + Back-end): Feature ensures that a scheduled query runs only when the queries dependencies have completed their run. Reduction in wasteful query runs lead to an approx 21% reduction of scheduled runs cost (this number also included some planned usage reduction activities which cannot be attributed to DAG)
• Maintained/stabilised the Notebook platform (custom integration of open-source JupyterLab)
• Architected and built a multi-tenant connector service for performing CRUD operations on external connection configurations
• ML Platform team (Ml Eng/ Mlops)
• A central self service platform to allow data scientists and analysts in the different stages of their model development life-cycle.
• Built and Maintained DS-Gateway service: Which allows other services in Udaan to easily make use of the deployed real time model services through a common thin client and allows doing things like model request response logging, shadow deployments, etc very easily
• Integrated Kedro with MLP (UI + Backend): Integrated Kedro (a python framework for writing maintainable and modular ETL pipelines) with MLP (V0 release) by building a UI layer to abstract creating and deploying Builds, access control, triggering and cancelling runs and viewing run logs. this product enabled 5 production use cases
• Was heavily involved in the planning, exploration and support stages (helping users onboard onto the platform) of Udaans ML engineering journey

Feb 2021 – Jul 2021 · 5 mos

• Worked on improving the autosuggest experience at udaan:
• Implemented a click popularity based ranking algorithm after building an offline POC by creating a pipeline to flow clicks data from events table to cosmos and then from cosmos to solr. The algorithm was launched as an A/B in electronics and clothing category, improvements in CTR (∼5% inc) , CVR (∼7% inc) and a decrease in avg query prefix length (∼5%) were observed

May 2020 – Jul 2020 · 2 mos · Bengaluru, Karnataka, India

• Tool for sorting images based on technical quality: Worked on building a command line tool which uses machine learning and deep learning techniques to assess the quality of an image and sort images based on the quality score(technical quality score)
• ML Algorithms: BRISQUE, PIQE, NIMA (neural image assessment)
• Data Set: KonIQ-10k, KADID-10k