Krupa Ajay Mehta

Deployed RAG platform integrating vector databases, semantic graph search, and hybrid retrieval, reducing query latency by 40%. Working on a Full-Stack AI Platform enabling scalable model hosting, serverless inferencing, and multi-tenant AI application deployment. Designed and implemented serverless inference architecture for hosting LLMs and vision models, optimizing cold-start latency and GPU utilization. Built production-grade Python backend services (FastAPI) for model orchestration, inference routing, request lifecycle management, and API gateway integration. Developed GPU-aware scheduling and monitoring systems for live GPU utilization tracking, VRAM monitoring, and performance diagnostics. Designed and deployed Agentic AI applications enabling autonomous multi-agent reasoning for enterprise workflows. Worked on performance optimization strategies, prefilldecode separation concepts, batching, GPU memory isolation, multi-instance scaling.

Built scalable backend infrastructure with comprehensive database transaction management and session handling. Implemented end-to-end AI solutions handling POC development, model fine-tuning with LoRA techniques, and production deployment. Familiar with C++ system-level programming: socket programming, inter-process communication (IPC), and OS-level. Developed and deployed real-time defect detection pipeline using PyTorch, and CUDA on Jetson and RTX 3060, 4070, and Ada series, achieving 30% faster processing. Collaborated with QA to resolve defects and led sprint planning and code reviews for a 3-member team.

Built and tested REST APIs for healthcare SaaS platforms. Tuned PostgreSQL and MongoDB for performance improvements. Worked with storing and processing large data and contributed to data visualization using Qlik View.