Software Engineer focused on systems programming, distributed systems, and ML infrastructure.
LinkedIn · GitHub · LeetCode · Blog · divi.verma1903@gmail.com
A GGUF-based LLM runner built from scratch. Custom tensor operations with multi-threaded matrix multiplication, memory-mapped file loading via Linux syscalls, byte-level GGUF parsing with custom 16-bit float structs, and RAII-based memory management for long-running processes.
A multi-provider JSON-RPC proxy with M-of-N consensus for data integrity. Includes a verification engine that walks Merkle-Patricia Tries and validates RLP-encoded proofs at the byte level, cryptographic state-root validation for Ethereum calls, and a real-time observability suite.
A peer-to-peer video conferencing platform built on WebRTC for low-latency communication. Establishes direct P2P data channels using UDP hole punching, and handles complex NAT traversal via a self-deployed and configured STUN/TURN server (coturn).
A Game Boy emulator with cycle-accurate CPU and MMU components. Handles memory-mapped I/O and hardware synchronization, renders 8-bit graphics via SDL2, and debugs system-level errors down to the register and instruction level.
Agentic Multi-Hop RAG Pipelines (in progress) — Benchmarking Repository Research paper on implementing multi-hop RAG pipelines with agent-based retrieval.
Tokenization in LLMs — Blog Implementing tokenization algorithms for LLMs from scratch in C++.
VIT Class of 2027 · Vellore, India