Most AI systems make the same mistake — every request goes to the same model. I built something that fixes that.
ai-gateway — AI Inference Gateway
A production-style AI inference gateway that routes LLM requests based on intent, model health, and cost.
- 14-layer request pipeline: rate limiting → auth → quota → cache → intent detection → routing → failover → tracking
- Embedding-based intent classification with LLM fallback
- Health scoring via Welford's algorithm — O(1) memory, no history stored
- Multi-tenant access control, Redis caching, observability dashboard
- 3 external contributors submitted PRs within 48 hours
- Deployed on Koyeb · Redis via Upstash · Inference via Groq · Infrastructure cost: $0
Inspired by a patent system I analyzed — routing logic from a UEI architecture became the foundation for intelligent LLM request handling.
Stack: Node.js Redis Express Groq Gemini Docker
Prompt2Visuals — Math Visualization Engine
Converts math prompts and equations into animations and plots for intuitive learning.
- Supports functions like
y = sin(x), linear equations, 2D/3D expressions - Real-time rendering pipeline — video or image output on demand
- Graceful fallback handling with domain validation
Stack: Python FastAPI Manim Uvicorn
Remote Collaboration Tool — Real-Time Platform
Full-stack collaboration platform with video, chat, AI summarization, and file management.
- WebRTC video calls via PeerJS, real-time chat via Socket.io
- AI transcription and meeting summarization using Google Gemini
- Built end-to-end: APIs, real-time systems, and frontend integration
Stack: Node.js Express MongoDB Socket.io PeerJS Google Gemini
Backend Node.js · Python · Express · FastAPI · REST APIs
Databases MongoDB · Redis
AI / LLMs Google Gemini · Groq · LLM Integration · Embeddings
Infra Docker · Koyeb · Render · Upstash
Auth JWT · OAuth
Real-time Socket.io · WebRTC · PeerJS
- SymPy — Fixed simplification for hyperbolic functions (
atanh(tanh(x))) - tldr-pages — Improved
kubectl-describedocumentation - freeCodeCamp — Corrected JavaScript curriculum content
- MCA @ Lovely Professional University (Graduating Aug 2026)
- Actively seeking: Backend Engineer / AI Engineer roles
- Open to: Remote · Delhi · Bangalore · Hyderabad · Mumbai
If you're building scalable backend systems or working on AI infrastructure — I'd love to connect.