Software Engineer · Full-Stack, Mobile & AI Systems · Karachi, Pakistan

• 💼 Software Engineer with 3+ years on systems that are actually in production, mostly Python (Django, FastAPI, Celery) and TypeScript (Next.js, NestJS, Vue.js).
• 🤖 A good part of my current job is the LLM side of a real-estate platform: routing requests across Anthropic and OpenAI model tiers, RAG and semantic search over ~50,000 documents a month, and counting tokens so I know what each feature costs to run.
• 📱 Shipped two React Native apps to the App Store and Play Store through an EAS Build/Submit pipeline.
• 🧰 I spend a fair amount of the week on the unglamorous half: Celery workers, queues, CI pipelines, and reviewing other people's code.
• 🎓 Computer & Information Systems Engineering, NED University. Two published papers.

self-healing-rag · A RAG service that critiques its own answers, refuses to guess, and can't be merged if it regresses.

A LangGraph loop where a critic model grades every draft against the retrieved chunks and sends ungrounded ones back for a reformulated query. If it still can't ground the answer, it abstains instead of guessing. Every push runs 60 hand-verified golden pairs through a cross-provider judge (Claude grading GPT output) and GitHub Actions blocks the merge if hallucination rate goes above 5%. There's an open PR on the repo sitting red on purpose, to prove the gate actually fires.

_{Current: 4.4% hallucination · 0% false answers on unanswerable questions · 100% valid citations · ~$0.0005/query}

🏆 3rd Prize, AI Track ($15,000), Solana Breakout Hackathon · out of 8,300+ global submissions, for Agent Arc, a non-custodial AI trading terminal on Solana where I led the frontend.

• Languages: Python, JavaScript/TypeScript, C#
• Backend: Django, FastAPI, Node.js, Express.js, NestJS, ASP.NET Core, Celery, REST, GraphQL
• Frontend: React.js, Next.js, Vue.js, Nuxt, Tailwind CSS
• Mobile: React Native, Expo, EAS Build/Submit, WebRTC, Socket.IO
• AI/LLM: RAG, LangChain, LangGraph, agents & tool use, embeddings & vector search, structured output, LLM evals, Whisper/TTS
• Data & Infra: PostgreSQL, MySQL, MongoDB, Redis, Docker, GitHub Actions, GitLab CI/CD, ETL & data pipelines

• An Integrated Approach to Robotic Object Grasping and Manipulation using 6D Pose Estimation · arXiv:2411.13205
• Smart Shelf Advertising using Real-Time Product Segmentation and Interaction on Low-Cost Edge Devices · DOI:10.13140/RG.2.2.34475.96800