Agents
Systems that decide
Multi-agent architectures with real tool use, hand-off between specialists, and a latency and cost budget that survives production traffic.
LangGraph · Pydantic AI · MCP · OpenAI · Claude · Gemini
AI Engineer · Lahore, Punjab, Pakistan · Open for work
Ali HassanAI Engineer
I build agentic AI systems, voice agents, RAG pipelines and computer vision models. Trained, deployed and left running in production. 150 repositories of it, in the open.
- Public repositories
- 150
- GitHub stars earned
- 298
- AI systems shipped
- 40+
- Client systems delivered
- 12+
What I actually do
A model in a notebook is a result. A model behind an endpoint, under load, with a fallback path, is a product.
Most of my work is the distance between those two things: labelling the data properly, choosing the smallest model that clears the bar, and building the service around it that a team can run without me.
Voice
Systems that talk
Agentic voice assistants that listen, reason and answer inside a turn. Live transcription, tool calls mid-conversation, speech back without the wait.
AssemblyAI · ElevenLabs · DeepSeek · Realtime APIs
Language
Systems that read
Retrieval over private documents, knowledge graphs where relationships matter, and federated setups for data that is not allowed to move.
LangChain · Vector databases · Neo4j · Hugging Face
Vision
Systems that see
Detection and recognition models trained on custom footage, then wrapped in a streaming inference service the camera can actually feed.
YOLOv11 · Ultralytics · OpenCV · MediaPipe
Selected work
Six that people kept using
Agents & LLMs
Agentic Stock Analysis
A multi-agent system where research, analysis and recommendation agents divide a market question between them and report back with reasoning.
Multi-agent pipeline
LangChainGroqPython6
Agents & LLMs
Voice AI Agent
A real-time voice agent wiring DeepSeek-R1 reasoning to AssemblyAI transcription and ElevenLabs speech, tuned for turn latency.
Real-time turns
DeepSeek-R1AssemblyAIElevenLabs2
Agents & LLMs
ResearchMind AI
A research assistant built on the programmatic agent hand-off pattern, so a question moves between specialist agents instead of one model guessing.
Agent hand-off pattern
Pydantic AIPythonOpenAI1
RAG & NLP
Knowledge Graph RAG
Production-ready retrieval that builds a knowledge graph first, so answers follow relationships in the source material rather than nearest-neighbour luck.
Graph-grounded answers
Neo4jLangChainPython1
Computer Vision
Face Emotion Detection
A YOLOv11 detector trained on a custom facial dataset that reads emotion straight from a live video feed, class by class, frame by frame.
5 emotion classes, real time
YOLOv11UltralyticsOpenCV31
Computer Vision
Smoking Detection
Workplace safety monitoring that flags smoking in real-time camera feeds, built for sites where a missed detection is a compliance failure.
Live video inference
YOLOv11OpenCVPython26
Available for work
Tell me what the model has to get right.
Send the problem, the data you have, and the deadline. You will get a straight answer about whether AI is the right tool for it.
Start a project