Portfolio

Umar Hassan

Senior AI/ML Engineer | 4+ Years Experience

Building production AI systems across core machine learning, Large Language Models (LLMs), Retrieval-Augmented Generation (RAG), and Agentic AI workflows.

View Projects Contact Me

Umar Hassan Umar Hassan Umar Hassan

Technical Skills

From AI foundations to Agentic systems

Results-driven Senior AI/ML Engineer with 4+ years of experience building, evaluating, and deploying production-grade artificial intelligence systems. Strong foundation in core AI and machine learning, extended into LLMs, RAG architectures, and Agentic AI. Focused on turning complex problems into scalable, reliable AI applications with Python, LangChain, and modern LLM tooling.

Languages and Data

Core programming and data stack for building AI systems.

Python (Pandas, NumPy, Matplotlib)
Scikit-Learn / TensorFlow / PyTorch
SQL
C++

AI & Machine Learning Foundations

Core AI/ML practice that underpins every production system I build.

Supervised learning
Model training & evaluation
Data preprocessing
ML pipelines
Feature engineering

Large Language Models (LLMs)

Building and evaluating LLM applications for reliable conversational and task-focused AI.

Prompt engineering
LLM application development
Response evaluation
Mistral / Groq / HuggingFace
Conversational systems

Retrieval-Augmented Generation (RAG)

Grounding LLM answers in private documents through retrieval and vector search.

Document ingestion & chunking
Embeddings
ChromaDB
Similarity / MMR retrieval
LangChain RAG pipelines

Agentic AI

Autonomous agents that plan tasks, reason over context, call tools/APIs, and complete multi-step workflows.

LangGraph multi-agent systems
Supervisor routing & guardrails
MCP tool integration
Human-in-the-loop workflows
LangChain agents
Tool / API calling

Computer Vision

Detection and tracking pipelines for real-world video data.

YOLOv8
OpenCV
Object detection
ByteTrack multi-object tracking

Deployment and Cloud

Shipping, hosting, and collaborating on production AI products.

AWS
GCP
Azure
Streamlit
Git / GitHub
joblib
Power BI / Excel / Tableau

Professional Experience

4+ years shipping production AI

End-to-end ownership as a Senior AI/ML Engineer across core ML, LLMs, RAG systems, and Agentic AI workflows.

Senior AI/ML Engineer

Lionic AI Labs · Lahore, Pakistan

Sep 2022 - Present (4 years)

Owned end-to-end delivery of production AI systems across a multi-year engagement — from core ML foundations through LLMs, RAG platforms, and Agentic AI workflows.

Built and continuously improved core AI/ML capabilities covering data preparation, model training, evaluation, and production deployment for real business use cases.
Designed and delivered Large Language Model (LLM) applications, including prompt design, response evaluation, and provider integration to improve answer quality, consistency, and latency.
Architected end-to-end Retrieval-Augmented Generation (RAG) systems that connect LLMs with vector databases for accurate, context-grounded answers over private and domain-specific documents.
Developed Agentic AI workflows using LangChain: autonomous agents with planning, multi-step reasoning, and tool/API calling to automate complex operational tasks.
Owned the full AI delivery lifecycle after release — monitoring quality, reducing hallucinations, refining retrieval and agent behavior, and partnering with stakeholders to scale solutions responsibly.

Projects

Work that shows the full stack

Detailed builds across multi-agent systems, RAG, conversational LLMs, and computer vision — each with source code on GitHub.

RAG System + Retrieval Evals

A production-style document Q&A RAG that retrieves evidence with multiple strategies, answers with source citations, and ships an evaluation table comparing dense, BM25, and hybrid retrieval head-to-head.

Full RAG pipeline: ingest (PDF/MD/TXT), chunking, local embeddings, Chroma persistence, and cited answers via free Groq (or Hugging Face).
Compares dense, BM25, hybrid (RRF), and hybrid+rerank retrieval with Hit@k, MRR, context recall/precision, and faithfulness metrics.
Shipped with a Typer CLI and FastAPI /ask endpoint so you can query documents and run strategy bake-offs from one codebase.

PythonRAGChromaDBBM25GroqFastAPIEvals

View Code

TripMate AI — Multi-Agent Travel Planner

A production-style multi-agent travel planner where specialized AI agents collaborate to search flights, hotels, and weather, assess budgets, and build itineraries — with supervisor routing, guardrails, and human approval before the final plan.

Orchestrates Flight, Hotel, Weather, Budget, and Itinerary agents with a LangGraph supervisor that selects only the specialists needed for each query.
Connects tools through MCP (Tavily, AviationStack, OpenWeather), enforces input guardrails, and pauses for human-in-the-loop approval before final synthesis.
Shipped with a FastAPI web UI, PostgreSQL/in-memory checkpoints for thread state, and a Docker deploy path for VPS hosting.

PythonLangGraphMCPGroqFastAPIAgentic AI

View Code

RAG Book Assistant

A document Q&A application that lets users ask natural-language questions over PDF content and receive grounded, context-aware answers powered by Retrieval-Augmented Generation.

Implements the full RAG pipeline: document loading, semantic chunking, HuggingFace embeddings, and persistent Chroma vector storage.
Uses similarity and MMR retrieval to pull the most relevant passages before generating answers with a Mistral LLM.
Shipped with a Streamlit interface for interactive PDF upload, querying, and transparent, context-grounded responses.

PythonLangChainChromaDBMistralStreamlitRAG

View Code

Mood-Adaptive AI Chatbot

An interactive conversational AI that adapts tone and personality in real time while retaining session memory — useful for more natural, context-aware dialogue experiences.

Dynamic system-prompt personality modes (for example calm, humorous, or assertive) that reshape response style on demand.
Session-based conversation memory so follow-up questions stay coherent across turns.
Delivered as both a CLI prototype and a Streamlit web app, with comparisons across Mistral, DeepSeek-R1, and Groq for quality and latency.

PythonLangChainMistralStreamlitLLMs

View Code

Sports Player Tracking Pipeline

A computer vision pipeline that detects and tracks players in sports video for downstream athletic analytics, producing annotated output with stable identities across frames.

Combines YOLOv8 person detection with ByteTrack multi-object tracking for persistent player IDs.
Uses OpenCV for frame processing, visualization, and export of fully annotated output video.
Designed for broadcast-style footage where reliable on-field tracking matters for movement and performance analysis.

YOLOv8OpenCVByteTrackPythonComputer Vision

View Code