Prasanna Kumar

Built a unified lakehouse and real-time ingestion foundation consolidating 500+ pipelines to a single governed source of truth. Established DataOps with >99.5% pipeline and platform uptime and standardized on-call and L1/L2 processes. Migrated compute to Kubernetes with spot instances, reducing cloud spend by 60% and improving throughput. Implemented enterprise data catalog, lineage, and PII encryption to meet RBI and internal audit mandates; achieved 100% audit compliance. Enabled standardized model onboarding and CI/CD, cutting deployment cycles by 50% and improving feature serving latency. Partnered cross-functionally to operationalize AI/ML decisioning, reducing manual underwriting and improving approval turnaround. Built a semantic layer to streamline BI access, increasing analyst efficiency by 30% and improving data trust via data contracts. Ingested high-volume event streams (1M+ daily calls) and delivered core operational metrics, improving lead conversion by 15%.

Led the architecture for low-latency blockchain monitoring to ingest and process billions of transactions in real-time. Built stateless Golang services on Kubernetes with distributed state for resilience and scale. Established streaming and batch pipelines for reliable landing in analytical stores and downstream analysis. Delivered internal tools that simplified heavy queries and accelerated investigations, reducing turnaround time. Integrated graph-based entity resolution to improve fraud detection accuracy by 30%. Reduced ETL batch processing time by 38% and cut resource usage, delivering faster data availability for analytics teams within 4 hours. Streamlined data lineage tooling, enabling analysts to complete regulatory reporting 25% faster, reducing external audit hours by 120 annually. Scaled data ingestion pipelines to handle a 3x peak load while maintaining 99.95% uptime, ensuring uninterrupted analytics access for teams globally. Raised test coverage to 92%, strengthened CI pipelines and reduced production incidents by 40%, improving deployment stability for stakeholders across all environments.

Designed and operated high-frequency pipelines to process market data in real-time across 200+ symbols. Implemented neural feature extraction with denoising autoencoders to enhance signal quality for downstream models. Delivered sub-second SLAs for critical components supporting time-sensitive operations. Hardened streaming jobs and monitoring to ensure consistent throughput and recoverability. Optimized ETL workloads through memory tuning cutting batch processing time from 12 minutes to 3 minutes for a measurable 75% efficiency gain. Scaled streaming topology to handle peak market data bursts by deploying partitioned processing and backpressure achieving 99.95% throughput without dropped messages. Implemented automated data quality checks and lineage tracing reducing production defects by 60% within 2 months and significantly stabilizing pipelines.

Architected a Kappa-based system ingesting 10,000+ beacons per second from 5,000+ IoT devices. Built pipelines with Kafka and Spark Streaming to transform data and persist into scalable stores. Enabled models for indoor positioning, movement analytics, and programmatic behavior insights for business stakeholders. Tuned topics, partitions, and schema strategies to support sustained growth and reliability.

Built data pipelines in Scala and Akka to convert logs into structured datasets for a Cassandra-based warehouse, enabling actionable insights.

Led the build-out of a Hadoop-based data lake for telecom analytics.

Developed Java applications using Spring MVC for legacy credit scoring and fraud detection on structured bureau datasets.

Java Developer on enterprise applications.