Experience

Problem

Engineering teams monitoring AWS environments often need to reason across multiple services, accounts, regions, metrics, dashboards, and logs before they can identify what happened and where to start troubleshooting.

Built / Contributed

Developed an observability platform that automated anomaly detection across multiple AWS services, accounts, and regions. Built Java backend services and a serverless metric ingestion pipeline using EventBridge, Lambda, and DynamoDB, collected metrics through cross-account IAM role assumption, integrated SageMaker for live time-series anomaly detection, and designed a triage interface with direct CloudWatch links to specific timestamps.

Problem

Data scientists needed a reliable way to deploy machine learning models to both Azure cloud infrastructure and NVIDIA Jetson edge devices without taking on container orchestration, hardware acceleration, and production API concerns.

Built / Contributed

Shipped a production API that enabled seamless ML model deployment to cloud and edge environments. Reduced inference latency by moving serving from CPU to GPU with TensorRT, adding in-memory model caching, and prewarming inference workers to eliminate cold starts. Integrated the platform into Altair's microservice architecture with RabbitMQ, Docker, and Kubernetes.

Problem

A study material chatbot needed faster responses and more reliable backend workflows for repeated vector queries, conversation history, context retrieval, and LLM orchestration.

Built / Contributed

Reduced API latency by implementing Redis caching for frequent vector queries, designed MongoDB schemas for conversation history, containerized backend services with Docker, and deployed them to Google Cloud Run to autoscale conversational workloads. Built automated integration and end-to-end tests with Pytest to validate context retrieval accuracy, database queries, and LLM orchestration workflows.