Services · Freelance & contract
AI engineering that ends in a deployed system, not a demo.
I am Ali Hassan, an AI Engineer working from Lahore, Punjab, Pakistan with teams anywhere. My work starts where the research paper stops: a model trained on your real data, wrapped in a service, running where your users are.
What I build
Six disciplines, one engineer accountable for all of them.
LLM & agents
Agentic AI & multi-agent systems
I build agents that use tools, call your APIs and pass work between each other, rather than one prompt trying to do everything. That includes MCP servers and the evaluation you need to know an agent is behaving.
Deliverables agent architecture, LangGraph orchestration, tool integrations, evaluation harness, cost and latency budget
Voice AI
Agentic voice chat agents
Voice assistants that listen, reason and answer inside a single turn, with tool calls happening mid-conversation. Built for support lines, booking flows and hands-free interfaces.
Deliverables realtime speech pipeline, agent logic, telephony or web integration, latency budget, call transcripts
Retrieval
RAG & document intelligence
I make private corpora answerable (contracts, tickets, manuals, research) with retrieval that gets measured rather than assumed. Where the data cannot leave its environment, I build knowledge graphs and federated setups that query it in place.
Deliverables ingestion pipeline, vector or graph store, retrieval evaluation, answer API
Computer vision
Computer vision systems
I train detection and classification models on your own footage, using YOLOv11 and custom architectures, and wrap them in pipelines that read live video without falling behind. That covers RTSP streams, edge devices and batch runs over recorded archives.
Deliverables dataset labelling strategy, trained model weights, evaluation report, FastAPI inference service, deployment
MLOps
MLOps & deployment
Most models stall between a working notebook and something a team can actually run. I containerise training and inference, track every experiment, and wire up the CI/CD and monitoring that let the next version ship without a rewrite.
Deliverables Docker images, AWS deployment, CI/CD, DVC and MLflow experiment tracking, monitoring, reproducible pipelines
Product engineering
AI product engineering
The model is rarely the whole job. Someone still has to log in, upload a file and read a result they trust. I build the application around the model, front to back, so you ship one thing instead of coordinating three contractors.
Deliverables Next.js or React front end, FastAPI backend, PostgreSQL, auth, dashboards, hosting
How it works
How an engagement runs.
01
Scope call
Thirty minutes to define the problem, look at the data you already hold, and agree what success will be measured against.
02
Proposal
A written approach with milestones and a fixed timeline, so you know what arrives and when before any work begins.
03
Build
Weekly demos running against your real data, so corrections happen in week two rather than at handover.
04
Handover
Code, model weights, documentation and a running deployment in your own accounts, with a walkthrough of how to operate it.
Ways to work together
Three formats. Pick whichever fits the work.
Fixed scope
Project
A defined build with an agreed scope, a fixed price and a delivery date. The right shape when you already know what needs to exist.
Ongoing
Retainer
A recurring block of my time for teams with models already running, covering retraining, new capability and the things that break.
Advisory
Consulting
A short engagement to review an existing architecture, choose between approaches, and give your team a direction it can build on.
Next step
Tell me what you are trying to build.
If it involves a camera feed, a model that never left the notebook or a pile of documents nobody can search, it is close to something I have built before.