01
Job Ad Classification at Scale
Six hundred million job adverts, sorted into standard occupations. The first working version would have cost $48,000 every time it ran.
AI Engineer / Lahore / Remote
I build AI systems that hold up in production.
Most AI work dies between the demo and the deploy. I do the part after the demo: the evaluation, the cost ceiling, the failure modes, the thing that still works on the six hundred millionth record.
What I build
Agentic workflows
50%
Automated Tariff ClassificationRAG on your data
<3.5s
Procurement Intelligence AgentVoice agents
<1s
Real-Time Voice CompanionDocuments & vision
$1–6
Reading Construction DrawingsData at scale
4×
Job Ad Classification at Scale
Selected work
01 — 0502
Conversation Analysis for Parents
A parent records a difficult conversation with their teenager and gets back what actually happened, why it mattered, and one evidence-based technique to try next.
03
Procurement Intelligence Agent
Buyers ask "what did we last pay for this, and to whom?" in plain language, and get the actual purchase orders back in about three seconds.
04
Reading Construction Drawings
Estimators price bids by reading plan sheets by hand. This reads the same sheets and lists every fence run, gate and height, with the sheet each came from.
05
Automated Tariff Classification
Classifying goods for customs was manual, slow, and impossible to staff at the rate the work arrived.
Also built
- Real-Time Voice Companion
- Real-Time Voice Sales Agent
- Multi-Agent CRM
- Thumbly
- AI Lawyer
- Anti-Spoofing Attendance
- Stock Exchange Data Platform
- Nutrition AI
Experience
5+ years / 6 rolesAssociate Data Scientist
Machine Learning Engineer
AI Developer
Associate AI Engineer
ML Engineer & Python Developer
Machine Learning Intern
About
Lahore, Pakistan, working remoteI work with teams who have an AI idea that needs to survive contact with real data, real budgets and real users.
That usually means one of three things: a retrieval or agent system that has to be accurate enough to trust, a pipeline that has to run at a scale where cost decisions matter, or a prototype that works on a laptop and needs to work in production.
I have done this for Pearson and PwC through TenX, and for a handful of startups building their first AI features. I write the evaluation harness before I write the pitch.
Working with
- Python · LangChain · LangGraph · FastAPI · PyTorch · Hugging Face
- RAG · FAISS · Qdrant · pgvector · Cohere Rerank · Multi-agent orchestration
- Databricks · PySpark · SageMaker · MLflow · AWS · Docker
- OpenCV · YOLO · MediaPipe · Deepgram · Cartesia · LiveKit
Education
BSc Computer Science
Contact
Tell me what you are building.
Useful things to include: what the system needs to do, what it runs on today, and what “working” would mean.
- Phone
- +92 332 070 2124
- Upwork
- Hire me on Upwork