Full-stack AI engineer building LLM systems, RAG pipelines, voice agents and computer vision that actually hold up in production. I care about the stuff that matters once real users show up, like cost, latency and reliability, not just a demo that looks good on launch day.
>
Built a production multi-tenant agentic commerce engine with Postgres row-level-security isolation, a WooCommerce adapter, a PII firewall and an owner dashboard, validated on a live 300+ product store. Added semantic retrieval with BGE-M3 embeddings on GPU and a cost-optimized LLM routing layer to keep per-conversation cost low at scale. Also built Knowlyst, a RAG system over web and video links, and benchmarked emerging LLMs on capability, latency and cost.
Built a self-hosted local LLM assistant for content & social media management; shipped text-to-image tools, automated HR screening and lead-generation agents; analyzed geospatial datasets for enterprise clients including CBL and EBM.
Built AI newsletter & blog generation tools with the Claude API, Flask and the YouTube Data API; automated team workflows and deployed scalable applications on AWS.
Developed Python scripts measuring 2D object distances in warehouse environments and applied open-source depth-estimation models to real-time CCTV feeds.
Built an Arabic-language medical voice assistant for Sahl AI — dataset prep, fine-tuning and integration — and ran comparative analysis of Wav2Vec vs. Google Speech for Urdu speech-to-text.
LLM observability & safety platform: model registry with risk classification, full audit logging, automated PII & toxicity detection, and one-click NIST AI RMF compliance reports. Published Python SDK on PyPI.
→
Production outbound and inbound voice agent for automated phone qualification and warm transfer. Multi-tenant, TCPA-compliant and backed by 586 hermetic tests, with a concurrency-safe dispatcher.
→
RAG assistant that lets legal teams query large PDFs in natural language with cited answers. Supports Gemini, OpenAI and local Ollama backends.
→
Offline backtesting harness validating trading strategies over ~9 years of BTC, ETH & SOL data, net of realistic fees and slippage — with strict no-look-ahead discipline and a go/no-go report.
→
Multimodal medical assistant that analyzes medical images, answers health questions, and supports voice interaction using LLMs and computer vision.
→
CrewAI multi-agent system that researches the market, plans content, generates visuals and writes captions to automate Instagram content end to end.
→20 projects · filter by domain
University of Karachi · 2021 — 2024
Final year project: SignVision, a real-time Urdu sign-language recognition system. Coursework in AI, NLP, statistics & data mining.
