TinyLlama Alpaca QLoRA
Instruction-tuned TinyLlama optimized with QLoRA for efficient deployment and lightweight inference workflows.
View model →I build practical AI products end-to-end: fine-tuned Hugging Face models, corrective RAG systems, and real-world multi-agent platforms. From model training to deployment, I focus on AI that delivers measurable outcomes.
Selected models I have trained and fine-tuned for real use cases in instruction following, reasoning, and medical AI.
Instruction-tuned TinyLlama optimized with QLoRA for efficient deployment and lightweight inference workflows.
View model →Reasoning-oriented fine-tuned Qwen variant designed for stronger thought-structured responses in complex prompts.
View model →Compact instruction model fine-tuned for low-resource experimentation and fast-turnaround prototyping.
View model →Domain-focused medical model for symptom understanding and condition-focused inference pipelines.
View model →Custom CNN pipeline built for medical image classification and disease-focused X-ray prediction tasks.
View model →Selected notebooks from my public Kaggle work focused on modeling, reasoning, and practical experimentation.
Notebook collection covering vLLM serving, diffusion pipelines, and GPU-accelerated inference workflows on AMD MI300X.
Two JavaScript MCP servers built to give LLMs live tooling access for market intelligence and developer workflows.
Multi-agent enterprise sales outreach system built for the AI Agent Olympics at AI Week Europe (Milan, 2026). Orchestrates a LangGraph pipeline of specialised agents — researcher, writer, and qualifier — to automate personalised B2B outreach at scale. Built with FastAPI, Groq (llama-3.3-70b), Tavily web search, and a React/Vite frontend. Supports PDF export and Gmail integration.
Advanced Islamic banking chatbot built on Corrective RAG (CRAG) — a self-verifying pipeline that crawls target websites, chunks and embeds content, retrieves relevant context, then runs a corrective layer that cross-checks generated answers against retrieved documents to eliminate hallucinations. Built for finance and banking use cases like policy lookups and FAQs.
AI-powered research tool that summarizes ArXiv papers, builds citation knowledge graphs, and lets you chat with any paper using RAG — built with FastAPI, HuggingFace, ChromaDB, and React with an animated glass-morphism UI.
A synthetic law firm knowledge base for training and evaluating RAG pipelines — covering civil litigation, criminal law, contracts, IP, employment, immigration, tax, bankruptcy, and legal ethics. Designed as a benchmark dataset for legal AI systems.
Web-based symptom checker where users describe symptoms in natural language or upload X-ray images to receive AI-powered condition predictions with confidence scores — returning top 1, 3, or 5 diagnoses via a backend NLP + image classification API.
Production chatbot powered by the Gemini API with full context management, streaming responses, and a clean conversational UI built with React and FastAPI.
Voice-activated AI assistant built on the OpenAI API that handles computer tasks, answers questions, and executes commands based on natural voice input.
Convolutional neural networks trained and fine-tuned for image classification, including custom architectures and transfer learning with PyTorch and TensorFlow.
Machine learning models trained to predict house prices based on various features and market trends using regression techniques and feature engineering.
Fine-tuned and published transformer models to the Hugging Face Hub — contributing to the open-source AI ecosystem with reproducible pipelines and PEFT/LoRA fine-tuning methods including LoRA, QLoRA, DPO, and GRPO.
Full-stack student management portal built for UK-based company GCRD — features student records, role-based authentication, and a full admin dashboard. Built with React, Express, and SQL. Deployed on VPS with Docker Compose and Nginx.
A call-centre application providing seamless communication and support for university students — built with a full MERN-style stack and deployed to production with automated CI/CD.
Full-stack task management application with user authentication, persistent storage, and a clean responsive interface built with React and Node.js.
Cross-platform food delivery app with real-time order tracking, user authentication, and a Firebase Realtime Database backend — built with Flutter and Dart.
Open to AI/ML engineering roles, research collaborations, and interesting projects. Based in Karachi, Pakistan.