Experience

Roles

• Roku

  September 2024 – Present

  Senior Software Development Engineer (Data) · Bangalore, India · Tech lead, 4 engineers reporting informally

  • Built an LLM-powered PII discovery and classification platform for the Ads Data Platform, replacing a planned six-figure annual vendor contract. It classifies schema metadata into sensitive-data categories and routes predictions to legal analysts for human validation.
  • Closed the loop between evaluation and data: clustered analyst corrections into six failure modes, generated provenance-tagged synthetic examples for the weak categories, and served them through retrieval-augmented few-shot prompting. Precision moved from 84% to 91% while holding recall at or above 99%.
  • Re-architected hourly Spark pipelines for ad impressions and advertiser conversions under a one-hour publishing SLA, replacing mutable tracker state with stateless, idempotent Iceberg processing built on partition-held state, Bloom filters, and positional deletes.
  • Added transactional write-audit-publish controls and safe reruns, cutting job runtime 35%, Trino query latency 20%, and incident recovery from a full day to two hours without increasing ingestion latency.
  • Architected a governed data-access layer that lets AI agents and engineers query roughly 5,000 Hive, Iceberg, and Snowflake tables through Trino, enforcing schema, lineage, ownership, identity context, and metric-certification policy before execution. Adopted by 120+ platform engineers.
  • Built an agentic Hive-to-Iceberg migration platform with mandatory human approval gates, CI/CD promotion controls, durable workflow state, and audit records across GitLab, Jira, Slack, and Confluence. Migrated 450 tables across 40 workflows with no production incidents.
  • Productionised Slack incident-history and Confluence runbook-retrieval MCP servers for an AI incident assistant, with read allowlists, least-privilege scopes, and human-gated writes. Deployed with Bazel, Terraform, and Kubernetes; adopted by 20+ teams and 150+ engineers.
  • Led the redesign of GDPR deletion across the data lake, introducing configurable compliance reporting and virtual-identifier rotation to immediately disassociate customer records. Cut end-to-end fulfilment from 90+ days to 5, a 94% reduction.
  • Re-architected GDPR data-download processing from MapReduce to Spark, scaling intake from 20 to 1,000+ requests per day while keeping daily physical reads tightly pruned relative to the source footprint.

• Groupon

  October 2020 – August 2024

  Software Development Engineer III · Bangalore, India

  • Designed and built a Java/Dropwizard batch and real-time personalization service supporting 40M emails and 60M push notifications daily at 5 ms p95 end-to-end latency.
  • Built a Consumer 360 data store with 400+ behavioral, demographic, and predictive features used by Marketing Engineering.
  • Led the architecture and phased cutover of 2,500 Spark jobs from on-premises infrastructure to Google Cloud Dataproc, designing bidirectional table synchronization to preserve marketing flows during per-job cutovers, then retiring the synchronization layer per channel without interrupting revenue paths.
  • Architected bulk-CRUD libraries for Amazon Keyspaces and Cloud Bigtable using batching and parallel execution, adopted as the standard access path by 10+ teams.
  • Led the migration of a high-throughput service datastore from Amazon Keyspaces to Cloud Bigtable with no customer-visible downtime, cutting its annual run cost 90% by replacing per-operation serverless pricing with a provisioned cluster.
  • Implemented retention and lifecycle policies in the data warehouse, reducing retained data volume by 80% and materially cutting annual storage cost.
  • Designed a data quality framework with record-, column-, and metric-level checks and alerting, reducing tracked data-corruption incidents by 75% over two quarters.

• AtkinsRéalis

  February 2019 – September 2020

  Software Development Engineer II · Bangalore, India

  • Built image and GIS data pipelines on Theia, a decoupled big-data and machine-learning platform on Databricks, and optimized PySpark workloads using Spark SQL and UDFs.
  • Built serverless ETL applications moving MongoDB data into Azure SQL Server for the Assist 2.0 workflow product.

• Eatclub

  June 2018 – January 2019

  Software Development Engineer I · Bangalore, India

  • Built backend APIs for vendor onboarding and inventory tracking, including QR-code integrations with Box8's e-commerce platform.

Skills

Languages

Python · Scala · Java · SQL

Data & distributed systems

Apache Spark · Trino · Kafka · Iceberg · Hive · Hadoop · Databricks · Airflow

AI data & governance

LLM data pipelines · Synthetic data · Retrieval-augmented generation · MCP · DataHub · PII classification · Provenance · Lineage · Policy-enforced access

Infrastructure & security

Terraform · Kubernetes · Bazel · OAuth · JWT

Cloud & datastores

GCP (Dataproc, Bigtable) · AWS (Keyspaces) · Cassandra · PostgreSQL · DynamoDB

Backend

Dropwizard (Java) · FastMCP / MCP (Python)

Education

• Indian Institute of Technology (Indian School of Mines), Dhanbad

  July 2014 – March 2018

  B.Tech, Minor in Computer Science and Engineering · CGPA 8.1/10

Internships

SNC-Lavalin Nuclear Solutions (2017) · AtkinsGlobal Digital Incubator (2016) · Saint Petersburg State University, Global Citizen Fellow (2015)