I build AI agents that plan, retry, and self-heal — and occasionally I do the same after a Messi highlight reel.
- 🎓 BS Artificial Intelligence @ FAST NUCES, Islamabad (2022 – 2026)
- 🤖 I specialize in multi-agent LLM systems, Agentic AI, and RAG pipelines — shipped a production 8-agent marketing SaaS and a self-healing RAG system that catches its own hallucinations (I wish TV writers had that feature)
- 👁️ Former Computer Vision Engineer Intern @ Camsense — built real-time CCTV surveillance pipelines (YOLOv8 + ByteTrack) pitched for deployment at AIR University & NASTP
- 🏅 IBM SkillsBuild certified: Generative AI Essentials — Using LLMs to Work with Data
- 💼 Open to AI Engineer / LLM / ML roles — my agents work autonomously, and so do I
| Project | What it does | Tech |
|---|---|---|
| 🎯 TakhleeqX | Production B2B SaaS: an 8-agent LangGraph pipeline that researches trends, writes bilingual (English/Roman Urdu) content, generates images, and renders MP4 reels — a full marketing team in one API call | LangGraph FastAPI React OpenAI |
| 🧠 Agentic RAG Codebase Assistant | Self-healing 11-step RAG with hybrid BM25 + vector retrieval (RRF fusion), cross-encoder reranking (+25% relevance), and Redis semantic caching (−30% latency) | LangChain Qdrant Redis Docker |
| 🏍️ Parking Lot Security System | Real-time CCTV theft detection on live footage — bike-owner matching, pose-based suspicion detection, auto evidence clips | YOLOv8 ByteTrack MediaPipe |
| 🌫️ FedHealth | Federated MLOps platform forecasting PM2.5 air quality across 9 distributed clients with differential privacy | PyTorch Kubernetes MLflow |
| 🏦 Cheque Verification | Document AI for automated bank cheque fraud detection with LayoutLMv3 + LoRA fine-tuning | LayoutLMv3 LoRA Streamlit |
| 🗞️ NLP Agentic Microservice Platform | AI-powered story generation microservice using Microsoft Phi-2 — fetches real-time trends and serves generated content over gRPC + FastAPI microservices | HuggingFace Phi-2 gRPC FastAPI Docker |
...and 30+ more experiments in the lab → all repositories
Languages & Core
AI / ML / LLMs
Backend, MLOps & Infra
- 🔧 Shipping production TypeScript at Tensor Labs × Huitzi Solutions (US) — data connectors, OAuth 2.0 flows, and incremental ETL on Google Cloud
- 🎙️ Building StreamVoice Agent — a fully offline voice AI: wake-word detection, streaming STT, sub-200ms TTS synthesis, zero cloud dependencies
- 📚 Going deeper on LLM evaluation frameworks and multi-agent reliability patterns
- ⚽ Still watching Messi highlights between builds — some things don't change
⚡ "Talent wins games, but teamwork and intelligence win championships." — and multi-agent systems win both.