Software Engineer at Metron Security (2025-09 – 2026-02)
Architected and built production AI pipelines and backend systems with focus on RAG and agentic systems.
- Architected multi-modal RAG pipeline processing 100GB+ documents (PDF, DOCX, CSV) with 95% accuracy(RAGAS), implementing intelligent chunking, AI-powered summarization for tables/images and vector embeddings.
- Built agentic AI system using LangGraph with dynamic tool selection, enabling users to query both internal documents and web sources through a single conversational interface.
- Developed async Ingestion pipeline with S3-presigned url for documents uploads; NextJS frontend featuring chat interface, document upload, ingestion pipeline visualization, configurable RAG strategies (basic/hybrid/multi-query).
- Engineered scalable backend infrastructure with FastAPI, Celery/Redis task queues, caching, and PostgreSQL with pgVector, RAGAs evals in CI pipeline and streaming responses. With latency 350ms-750ms + Streaming response.
- Optimized Commander's Service Mode cloud vault sync calls across 15+ commands, cutting response times by ∼40% — improving reliability for app integrations like Slack.
Software Engineer at Cognizant Technology Solutions (2022-08 – 2025-07)
Built scalable backend systems, microservices, and AI pipelines for enterprise clients including Australia's largest telecom.
- Implemented session-aware RAG Q&A through AWS Lambda and Bedrock Knowledge Base embedded with Amazon Titan Text Embeddings v2, so follow-up questions remain in context.
- Delivered grounded answers via retrieve-then-generate, with source citations when returned—rather than LLM-only replies.
- Engineered Order & Cart Microservices for Telstra's 5G network rollout — Australia's largest telecom — across B2B and B2C flows, implementing orchestration-based Saga pattern for order consistency, compensating transactions, and failure recovery.
- Architected async Excel/PDF report generation and email delivery pipelines using RabbitMQ, ensuring background job processing, decoupled delivery, and zero report loss on service failures.
- Optimized API performance via schema denormalization and targeted column indexing reducing response time by 30%; cut image size 50% using multi-stage Docker builds, reducing CI/CD pipeline duration.
