Basant

Owned the App Analytics Agent Platform end-to-end: architecture on Google-ADK, LangChain, and CrewAI with MCP servers for agent-to-agent communication and LangGraph orchestration. Platform runs autonomously across 500+ mobile apps, cutting analytics delivery time by 60% and analyst manual effort by 75%

Addressed a video analysis bottleneck by building RAG pipelines on Milvus and integrating LLM-based processing for 1M+ videos/month, resulting in 20% higher engagement insights. Separately deployed ReAct agents for funnel analysis and churn prediction

Established production observability for all LLM agents through LangSmith, LangFuse, and Opik, enabling full trace visibility into agent decisions and prompt-level regression detection

Manage Databricks + Unity Catalog infrastructure handling 10TB+ daily, including the data governance layer, lineage tracking, and schema evolution

Drove MLOps adoption with MLflow: recommendation system increased personalization by 25%. Also maintain serverless microservices processing 1M+ events/second on Lambda

Collaborate with product managers, data scientists, and business stakeholders to define data strategy, prioritize pipeline work, and ensure data security and access control standards are met

Mentor 5 engineers through structured code reviews and cross-team pairing sessions between data engineering and ML. Team delivery times decreased 30%

Rebuilt the core ETL layer in Spark and PySpark from the ground up. Processing time for 5TB+/day of mobile analytics data dropped 50%, which unblocked the product team on real-time dashboards

Tuned distributed databases across Trino, DuckDB, CitusData, TimescaleDB, and ClickHouse serving 100M+ queries/day. Query optimization and repartitioning delivered a 50% speed improvement

Led Kubernetes migration of the full data infrastructure. End result: 40% cost reduction, 99.9% uptime, and auto-scaling that held under peak load

Designed the Iceberg-based Lakehouse with Unity Catalog and SCD patterns, now handling petabyte-scale data at 35% lower storage cost

Built Kafka + Kinesis streaming pipelines feeding real-time analytics for 500+ apps, alongside a data quality framework maintaining 99.5% accuracy

Owned data modeling for analytics use cases and automated pipeline orchestration using Airflow, reducing manual intervention and improving scheduling reliability

Operated all production workloads on AWS (Lambda, EC2, RDS, EKS, S3, Redshift, Glue, Athena) with monitoring and alerting configured from initial deployment

Led a team of 4 developers on an Intelligent Dialogue System. NLP + deep learning approach achieved 90% intent recognition, meeting the client acceptance threshold

Coordinated requirements gathering, sprint planning, and delivery milestones across academic and industry stakeholders. Project completed two weeks ahead of schedule