AI Engineer | Building reliable LLM, computer-vision, and data products
I build AI systems beyond the model: the data pipelines, APIs, evaluation workflows, validation rules, and interfaces that make an AI feature dependable in real use. My focus is applied machine learning, LLM applications, retrieval systems, and computer vision.
Currently: AI Engineer at MAAKWEB (part-time, remote) and a BS Artificial Intelligence student at FAST-NUCES (2022-2026).
A portfolio-grade enterprise knowledge assistant in active development. CiteGuard combines citation-first answers, hybrid retrieval, role-based access control, provider fallback, grounding checks, and reproducible evaluation.
Engineering focus: permissions are enforced inside the vector database query, unsupported answers are refused, and performance claims will be based on measured evaluation results rather than assumptions.
Python FastAPI Streamlit LangGraph Qdrant Docker GitHub Actions
An invoice and receipt-processing system designed for automated bookkeeping,
with reliability as the primary goal. DocFlow combines OCR and vision-language
extraction with deterministic Decimal arithmetic checks, self-correction,
confidence scoring, and human-review routing.
What makes it different: auto-approval is earned through validation rather than assumed from a model response. The project is being built with a strict dev/test split so results remain reproducible and honest.
Python FastAPI Streamlit Pydantic LangGraph PostgreSQL OpenCV Docker
Final-year project that analyzes football kicking technique from video and produces personalized coaching feedback. It uses computer vision for movement extraction, Dynamic Time Warping for motion comparison, and an agent workflow to explain the result.
Python YOLOv8 MediaPipe LangGraph FastAPI PostgreSQL Docker
Forecasting system for Pakistani commodity prices, from data collection and cleaning through model comparison with XGBoost and Random Forest.
Python Pandas XGBoost scikit-learn
| Area | Tools and methods |
|---|---|
| LLM systems | LangGraph, LangChain, structured outputs, prompt design, tool use, evaluation |
| Retrieval | RAG, embeddings, hybrid search, reranking, Qdrant, FAISS |
| ML and vision | PyTorch, Hugging Face, YOLO, OpenCV, MediaPipe, scikit-learn |
| Backend and data | Python, FastAPI, PostgreSQL, REST APIs, data pipelines |
| Delivery | Docker, GitHub Actions, testing, reproducible workflows |
- Security is structural: permissions should be enforced by the system, not requested from a model.
- Evaluation before claims: results belong in reports only after the scripts have produced them.
- Visible uncertainty: when evidence is insufficient, the product should say so clearly and guide the user to the next step.
- Explainable systems: every important architecture choice should be easy to defend in an interview or a design review.
I am building CiteGuard and DocFlow into complete portfolio-grade systems that show reliable AI engineering from data preparation through deployment, monitoring, and evaluation.
I am open to remote AI Engineer, ML Engineer, and LLM Engineer roles. Reach out through LinkedIn or email.