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

> Initializing profile...
> Loading projects: [AgentOps] [Orion] [AvatarAI] [SkinCancer]
> Status: Building in public · Shipping products · Learning daily

I build AI infrastructure — not demos. Right now I'm deep in multi-agent orchestration, real-time desktop AI assistants, and 3D avatar systems. I care about systems that actually work: fast pipelines, clean architecture, and production-grade code.

75+ contributions  ·  50+ commits  ·  3 production apps deployed

• Architecting multi-agent LLM pipelines with LangGraph
• Shipping Electron desktop apps with AI vision & auto-update
• Building real-time 3D avatar systems powered by MediaPipe
• Designing RAG memory systems at scale (ChromaDB + Neon Postgres)
• Running AI provider waterfalls across Groq, Cerebras, Gemini, SambaNova

A production-grade orchestration system where five specialized agents collaborate to solve complex tasks in real-time.

Supervisor → Planner → Researcher + Coder + Critic
         ↓
   WebSocket Stream → Redis Pub/Sub → Frontend

Stack: FastAPI LangGraph Neon Postgres Redis ChromaDB Docker Next.js Supabase

What it does:

• 5-agent diamond pipeline with supervisor routing and critic feedback loops
• Per-user RAG isolation in ChromaDB with semantic memory retrieval
• Real-time WebSocket streaming via Redis pub/sub
• Docker sandboxed code execution environment
• GitHub automation and file management agents
• Google OAuth + rate limiting with slowapi

A floating AI copilot that lives on your desktop. Understands your screen. Answers instantly.

Screenshot → LLaVA (local OCR) → Provider Waterfall → Response
Ctrl+Space ──────────────────────────────────────────────────→ Overlay

Stack: Electron Ollama LLaVA FastAPI Neon Postgres Groq Cerebras Gemini

What it does:

• Global Ctrl+Space hotkey — zero friction activation
• Local screenshot extraction via Ollama LLaVA with optional cloud reasoning
• AI provider waterfall: Groq → Cerebras → Gemini → SambaNova → Together → HuggingFace → Ollama
• Persistent conversation memory in Neon Postgres
• Auto-update via electron-updater, draggable always-on-top overlay
• CSP-hardened renderer, XSS-safe output

Your face, reconstructed in 3D and animated in real-time using only a browser camera.

Camera → MediaPipe FaceMesh + Pose → React Three Fiber → Three.js Avatar
                ↓
       Emotion Detection → Avatar Expression Sync

Stack: React Three.js React Three Fiber MediaPipe PostgreSQL Prisma Vercel

What it does:

• Real-time FaceMesh (468 landmarks) + full body pose estimation
• Emotion detection pipeline driving avatar expressions
• Fully customizable 3D avatar system
• Persistent user data via PostgreSQL + Prisma ORM
• Zero native dependencies — runs entirely in the browser

{
  "languages":   ["Python", "TypeScript", "JavaScript"],
  "backend":     ["FastAPI", "Node.js", "PostgreSQL", "Redis", "Prisma"],
  "ai_ml":       ["LangGraph", "LangChain", "ChromaDB", "Ollama", "MediaPipe"],
  "ai_apis":     ["Groq", "Cerebras", "Gemini", "SambaNova", "Together AI"],
  "frontend":    ["Next.js", "React", "Three.js", "Framer Motion", "Tailwind"],
  "desktop":     ["Electron", "electron-updater", "electron-builder"],
  "infra":       ["Docker", "Neon", "Supabase", "Railway", "Vercel"],
  "tools":       ["Git", "GitHub Actions", "Redis Pub/Sub", "WebSockets"]
}

[ AgentOps ] ──── [ Orion ] ──── [ AvatarAI ]
     │                │                │
  FastAPI          Electron         Three.js
  LangGraph        Ollama           MediaPipe
  ChromaDB         Groq             Prisma
     │                │                │
     └────────────────┴────────────────┘
              Built by Vansh