██████╗ ██╗  ██╗ █████╗ ███╗   ██╗██╗   ██╗███████╗██╗  ██╗
██╔══██╗██║  ██║██╔══██╗████╗  ██║██║   ██║██╔════╝██║  ██║
██║  ██║███████║███████║██╔██╗ ██║██║   ██║███████╗███████║
██║  ██║██╔══██║██╔══██║██║╚██╗██║██║   ██║╚════██║██╔══██║
██████╔╝██║  ██║██║  ██║██║ ╚████║╚██████╔╝███████║██║  ██║
╚═════╝ ╚═╝  ╚═╝╚═╝  ╚═╝╚═╝  ╚═══╝ ╚═════╝ ╚══════╝╚═╝  ╚═╝

dhanush = {
    "role"      : ["AI Engineer", "Full-Stack Developer", "Backend Architect"],
    "location"  : "Tamil Nadu, India",
    "focus"     : ["Multi-Agent Systems", "LLM Security", "Distributed Backends"],
    "currently" : ["Building production-grade AI systems", "Open Source LLM tooling"],
    "open_to"   : ["AI Engineer", "Full-Stack", "Backend", "LLM Engineer roles"],
}

AI Engineer and Full Stack Developer specializing in multi-agent AI systems, LLM orchestration, and real-time distributed backends — combining solid engineering fundamentals with cutting-edge AI to ship products that matter.

🤖 AI & GenAI

🖥️ Frontend

⚙️ Backend

🗄️ Databases

🚀 DevOps & Cloud

🧠 Core

• Built 5-agent LangGraph system — parallel Search Swarm + RAG Vault — dynamic routing to arXiv, PubMed, GitHub, Tavily — only ≥7/10 Critic-scored reports reach users
• Engineered RAG pipeline — PDF, Word, YouTube, URL ingestion — Gemini embeddings into scoped Qdrant collections — vector retrieval under 80ms with Groq → Gemini failover chain
• Developed custom MCP server exposing Tavily + Qdrant as standardized tools — modular tool-calling across all 5 agents via unified MCP protocol
• Streamed agent output via SSE — 7 event types — first token under 3.5 seconds — PostgreSQL persistence, Redis memory and 24-hour semantic report cache
• Implemented security layer — jailbreak mitigation, rate limiting 3 req/min, input sanitization — 5-table PostgreSQL telemetry tracking token count and quality per agent

Cloudflare WAF — but built specifically for LLM applications

• Built production LLM Security Gateway — API Gateway, Input Guard, Output Guard, LLM Proxy, Audit services each independently deployable via Docker Compose
• Implemented prompt injection detection, jailbreak classifier and PII scrubbing via Microsoft Presidio — blocked 100% known attack patterns under 50ms latency overhead
• Built output moderation pipeline — toxicity detection via HuggingFace Detoxify, hallucination check and format validation before every LLM response
• Designed YAML-based zero-config setup — fully customizable with plugin system for custom validators — single Docker command self-hostable
• Built real-time threat dashboard — blocked requests, PII frequency, threat analytics — per-API-key audit logs and instant key revocation

• Built production-grade distributed job queue from scratch in Node.js + TypeScript replicating BullMQ-level internals with full architectural transparency
• Implemented all state transitions — enqueue, move-to-active, fail, stall recovery — as atomic Redis Lua scripts eliminating race conditions across horizontally scaled workers
• Designed distributed sliding-window rate limiter, heartbeat/watchdog stall detection and exponential backoff retry using Redis Sorted Sets for high/normal/low priority
• Built real-time React dashboard with Socket.IO streaming and p50/p95/p99 latency percentile tracking across full REST API control plane

✅ 1000+ problems solved across LeetCode & GeeksforGeeks
🏆 Rank #1 GeeksforGeeks College Coding Score
⭐ Open Source Contributor — Appwrite, n8n, Hoppscotch
📊 2,013 GitHub contributions in last year — 76+ public repositories
📝 Technical blogs published on Medium
🚀 30+ full-stack projects shipped across AI, Web and Infrastructure