I'm an AI/ML Engineer with 3+ years of experience building and shipping production-grade LLM systems at scale. I specialize in RAG architectures, Voice AI, distributed inference, and real-time conversational agents — turning cutting-edge research into systems that actually work in the real world.
- 🏗️ Architected distributed LLM inference platforms serving 1M+ daily queries on AWS EKS
- 🎙️ Built multilingual Voice AI agents using LiveKit, WebRTC, Twilio, OpenAI & Deepgram — automating 50K+ calls/month
- ⚡ Reduced LLM inference latency 50% (400ms → 200ms p95) and costs 30% via GPTQ quantization & vLLM
- 📉 Cut hallucinations 40% using hybrid Graph + Vector RAG on enterprise knowledge bases (10M+ docs)
- 🌍 Published low-resource NLP research for Urdu (70M+ speakers) — targeting ACL/EMNLP 2025
| Project | Stack | Highlights |
|---|---|---|
| AI Calling Agents | LiveKit, WebRTC, Twilio, OpenAI, Deepgram | 100K+ users, 95% automation, 50+ business customers |
| Call Analytics SaaS | FastAPI, LLM, Dashboards | Real-time sentiment, summaries & conversation analytics |
| Dental AI Bot | RAG, WhatsApp/Instagram/Twitter | Domain-specific omnichannel chatbot |
| Restaurant AI Bot | Multimodal RAG, Pinecone, LangGraph | Multi-platform reservations, menu image understanding |
| ICAP AI Bot | LLaMA 3.1, RAG, Ubuntu On-Prem | On-premises deployment, enhanced data privacy |
| Sports Commentary AI | OpenCV, GPT-4, TTS, Edge Computing | Live cricket commentary, sub-second latency |
| Urdu NLP Research | Gemma, PEFT/QLoRA, Custom TTS | Fine-tuned for 70M+ Urdu speakers, targeting ACL/EMNLP 2025 |
🎓 BS Software Engineering — University of Karachi (UBIT) · 2020–2024
📜 Certifications:
- LLMOps · Agentic RAG with LlamaIndex · Pretraining LLMs · Prompt Engineering · Building Systems with ChatGPT — DeepLearning.AI
- Introduction to Generative AI — Google Cloud
- Machine Learning · Deep Learning · Notebook Expert — Kaggle
- n8n Automation — Simplilearn
- Intermediate Python — DataCamp
- 📄 Urdu TTS Architectures — Novel approaches to natural-sounding speech synthesis for low-resource languages
- 📄 Efficient LLM Adaptation for Low-Resource Languages — PEFT/QLoRA techniques for Urdu and similar languages
Targeting ACL / EMNLP 2025
