Software Engineer (Distributed DB & Query Systems, Research)

Aug 2025 - Nov 2025

• Optimized query execution via ML models (DQN, SVM) to predict operator costs and guide query planning, reducing p95 latency by 25%.
• Replaced coordinator-based ingest with per-instance sharded I/O (Parquet via Arrow C++), achieving up to 1.85x parallel ingest throughput.
• Validated system under concurrent query workloads using Spark over 10M records and managed ML model lifecycle with TensorFlow Serving.

Tech: SciDB (RocksDB-backed), PostgreSQL, Python, C++, TensorFlow Serving, Spark (PySpark), Arrow, Parquet

Software Engineer (Distributed DB & Query Systems, Graduate Research)

May 2024 - Dec 2024

• Built a C++ query rewriting layer that parses native DB queries and canonicalizes them to generate semantically equivalent variants.
• Enabled multi-tenant query isolation via a cloud-native control-plane gateway (AWS EKS) with SLA-aware routing and per-tenant rate limiting.
• Built a PostgreSQL feature store for training and snapshots, with Prometheus and Grafana dashboards for system-level metrics.

Tech: SciDB (RocksDB-backed), PostgreSQL, Python, C++, Prometheus, Grafana, Docker, AWS (EC2, RDS, EKS)

Backend Integrations

• Built event-driven publishing service (Spring Boot + Kafka) with backpressure-aware throttling, improving editorial throughput by 30%.
• Added social-media publishing (REST APIs + OAuth 2.0) with automatic fan-out and quota control, driving engagement by 20%.
• Improved real-time publish visibility by speeding DB writes (Hibernate batching, connection pooling) and CDN edge caching with auto-purge.
• Added unified observability with Prometheus + Grafana and fault-injection tests (Chaos Monkey), reducing MTTR by 20%.

Tech: Java, Spring Boot, Kafka, Hibernate, Apache DBCP2, Akamai CDN, JMeter, Prometheus/Grafana, Resilience testing

Platform Infrastructure

• Built concurrent data ingestion pipelines (Java ExecutorService), shifting from serialized to batch processing for 35% throughput improvement.
• Developed high-speed Java servlets for data transformation and schema validation with consistent sub-50 ms/file latency.
• Designed a JMS-based transaction monitoring service for reliability and exactly-once delivery across concurrent ingestion pipelines.
• Added DLQ replay and transactional messaging for fault recovery, achieving 99.9% uptime and supporting 3K+ daily feed transactions.

Tech: Java EE, Servlets, JMS (ActiveMQ), Apache Saxon, JMeter, JVisualVM, JUnit, Multithreading