Bhanunjay K.

20+ years of commercial software engineering combined with 19 years of critical technical maintenance in mission-critical systems within the Indian Air Force. I do not build basic API wrappers or simple demo applications. I engineer production-grade enterprise infrastructure, isolated local language model clusters, and high-throughput distributed microservices designed to maintain flawless reliability under intense multi-user scale. How I can assist in your engineering project s: 🔒 1. Sovereign & Air-Gapped AI Architecture I specialize in deploying, fine-tuning, and optimizing open-weight LLMs (Qwen, Llama) within strict, fully isolated air-gapped environments. By leveraging frameworks like vLLM and writing custom PyTorch CUDA extensions, I mathematically fit advanced model layers into constrained local hardware (e.g., 24GB VRAM nodes) without runtime memory crashes or quantization drift. I replace fragile prompt engineering with deterministic token-level grammar controls (CFG) to guarantee 100% reliable structural outputs. 🤖 2. Governed & Deterministic Agentic Ecosystems I construct resilient multi-agent pipelines incorporating error-recovery fallback nodes, circuit breakers, and GraphRAG architectures. If you need secure multi-agent orchestration that handles real-world codebase automation, loop limits, and complex operational logic with zero prompt drift, I deliver it. ⚙️ 3. High-Concurrency Backend & Low-Latency Infra I design low-latency distributed microservices and resilient API gateways utilizing Go (Goroutines, Channels), gRPC, and WebRTC. I have spent 4 years building high-throughput, real-cash interactive gaming environments requiring ultra-low latency data synchronization. 📊 4. Computer Vision, Deep Learning & Real-Time Analytics I build deep learning pipelines and custom PyTorch CUDA extensions that yield 200x–500x compute speedups (e.g., native C++/CUDA inference integrations linking Reinforcement Learning agents running natively in high-fidelity Unity simulations). 🔒 5. Hardened Linux & IoT Security My expertise extends directly to the kernel level. I build hardened Ubuntu IoT gateways featuring X.509/mTLS identity management, signed A/B OTA updates, and outbound WireGuard tunnels. 📊 6. Multiplayer Games: I have developed 300+ multiplayer games including real cash casino games, Slot games, Blackjack, Bingo, Plinko games. Tech Stack: Reactjs, phaser3, Threejs, Nodejs, Go, socket io, mongo db. 🚀 FEATURED PORTFOLIO HIGHLIGHTS: * Construct-lite (Governed AI Codebase Agents): Engineered a self-hostable gateway routing requests across local and cloud LLMs for governed multi-agent workflows. Takes a GitHub issue through automated planning, coding, review, and a human approval gate to open a PR. Built with fallback chains, circuit breakers, per-run token budgets, loop limits, end-to-end tracing, and prompt-injection screening. Deployed as 7 microservices (Go/Python) with Docker Compose using a fully free open stack: Ollama, Postgres, Redis, NATS, LangGraph, MCP, and OpenTelemetry. * StreamGate (Go Gateway & WebRTC Engine): Engineered an enterprise-scale distributed routing layer managing over 18,000 concurrent WebRTC live streams with a packet-loss threshold below 1% using Go microservices, gRPC backplanes, and SFU media configurations. * Computer-Vision Snooker Referee: Developed a real-time deep learning computer vision system combining Python, OpenCV, and a Phaser 3 game client to identify ball vectors and trajectories as a low-latency automated rule referee engine. * AI Character Animation Pipeline: Devised a custom full-stack pipeline triggering deep generative learning models via frontend controls to drive dynamic automated vector character frames. * Signal Audit Lab: Developed an advanced metrics monitoring suite designed for high-velocity signal auditing, deep hardware metrics tracking, and real-time network parameter visualization. * HyperDrive-RL Plugin: Optimized native C++/CUDA inference integration linking Deep Reinforcement Learning agents to high-fidelity Unity simulations, reaching up to 500x acceleration. Let's connect to discuss your project's architectural constraints, security guardrails, and performance requirements.