Ved parkash Soubhari

I’m pursuing M.Tech (AI) at the Indian Institute of Technology, Patna, and I build systems that turn models into usable products. My work centers on performance, concurrency, and practical integration across inference, APIs, and user interfaces.

In my recent projects, I built a multithreaded LLM inference runtime for efficient local AI serving. I implemented a thread-safe request queue, HTTP server, and FCFS/Priority/SJF scheduling, plus dynamic batching and an LRU KV cache for concurrent inference using C++17 and CUDA, integrated with FastAPI and Streamlit via llama.cpp.

I also developed a Deep Packet Inspection engine from PCAP captures by extracting connection metadata and implementing TLS SNI extraction for application identification in encrypted HTTPS traffic. Earlier, I completed a React internship (reusable UI components, REST API integration, state management) and a MERN full-stack training with authentication, routing, and deployment on Netlify and Vercel.