Project 1 — Microservices-Based Hybrid Retrieval Platform with Distributed AI Document Ingestion

• Implemented MCP (Model Context Protocol) using Azure Function Apps with optimized real-time Flask SSE streaming on OpenShift; added an intelligent latency-aware caching layer cutting per-request setup time from 6s → 1.6s (73% faster), improving grounding & response latency by 50% over traditional RAG.
• Built a production-scale event-driven ingestion pipeline (Azure Functions, Service Bus, AI Search) with queue-based orchestration & stage-level retries, processing 1M+ documents while cutting pipeline cost by 25%.
• Optimized vector search with pre/post-filtering in Azure AI Search — +30% retrieval precision, -50% latency; built a Tesseract OCR pipeline with 90%+ accuracy.
• Upgraded GPT-4o mini → GPT-5.4 mini and ada-002 → text-embedding-3-large (1536D) for better accuracy & hallucination resistance.
• Engineered correlation ID-based distributed tracing & Dynatrace observability dashboards across microservices.

PythonAzure FunctionsEvent GridService BusMicrosoft FoundryDoc IntelligenceRAGMCPAzure AI SearchTesseract OCRFlask (SSE)GunicornDynatraceSonarQubeOpenShiftGitHub Copilot

Project 2 — GenAI Engineer: HPC Modernization (Fortran → Cloud)

• Migrated a legacy Fortran on-prem DLL engine to AWS Cloud, optimizing core modules in C++/Python for 2–3× scalability & 5–10% faster execution.
• Implemented OpenMP/MPI-style parallel processing while preserving benchmark parity with the legacy implementation.
• Migrated on-prem databases to AWS achieving 99.9% availability.
• Used Claude for code understanding & refactoring, cutting dev effort by 70%.

C++Python (NumPy)AWS (EC2, Terraform, CI/CD, Redis, Load Balancer)OpenMP/MPISnykClaude Code

High-Performance Workfeed System

• Designed a scalable backend validated via Autocannon load testing, sustaining 50k+ concurrent users & 100k+ requests/min.
• Improved horizontal scalability with PostgreSQL sharding, read replicas, Redis caching & S3 storage — enabling 200× user growth.
• Built a video ingestion pipeline (multipart uploads, FFmpeg, AWS Lambda, MediaConvert) for adaptive encoding up to 4K.
• Implemented an API Gateway + event-driven microservices architecture for fault isolation & async processing.

React NativeNode.jsSupabaseAWS (EC2, S3, Lambda, CloudFront, MediaConvert)RazorpayFFmpegBullMQ/RedisNativeWindDockerKubernetes

🔧 Work in progress: Context API → Redux migration, Supabase monolith → AWS microservices.