Zaveen00
Pre-game2026
00The odds
Portrait of Zaveen Waqar

Head of AI, Compact Machines

I build AI that has to work outside the demo: in diagnostic labs, on factory floors, in classrooms and in M&A datarooms.

This is my story told as a basketball game, because that's how I lived it. Missed shots, one big rebound, and a long drive up the court. Five games along the way, each built on something I actually shipped.

Start the story

ZaveenWaqar

EE ends the CS department's title streak Search-and-rescue UGV wins at FIT 2016 Same robot takes CEPEX Open House 2017 Fabric inspection goes live on NVIDIA Jetson TX2 Morph-AI, OCR-AI and FISH-AI cut pathology turnaround Waqar named Director AI at AILA Waqar joins Compact Machines as Head of AI

The long way up the court

Here's the whole route before we play it out. Dashed lines are the shots that didn't fall. Solid lines are the moves that did.

Tap a spot on the court to jump to that part of the story.

  • Moved
  • Dribbled
  • Took a shot
  • Missed
UK: no way to payNUST: by a hairTsix-month timeout1City School2A-levels3COMSATS4Robotics5NUST MS6HyperNym7STech8siParadigm9AILA10Compact Machinesbasketball over booksthe reboundthe pivotfull circleLahorefive seasonsand counting

Tip-off

The City School

I did every year of school at The City School. I was a regular student with one real edge: I spoke English well, and that was entirely my mother's doing.

Assist: Mom

The miss

2010 to 2013

Odds +2

Somewhere in O-levels I picked up a basketball and didn't put it down. By A-levels I'd joined a club, my evenings belonged to the court, and my grades paid for it.

Odds +2

I applied to universities in the UK and got into two. We couldn't afford either.

Odds +3

So I applied to NUST. Being a complete novice, I applied for exactly one discipline. I missed the cut by a hair, and with no second choice on the form, there was nowhere else to go.

Timeout

Six months later I walked into COMSATS. It wasn't my first choice. It was supposed to be a stopover.

0 for 3 that year

The rebound

2013 to 2015

The plan was to reapply to NUST the next year. Then I found out COMSATS had a serious basketball scene, and the plan quietly died.

Zaveen +2

For two years, basketball came first and classes came a distant second. I played at HEC level, on the national inter-university circuit.

Zaveen +3

The computer science department had owned the inter-department title for years. I led Electrical Engineering to the title and ended their streak.

Zaveen +3

Then we did it again the next season. First lead of the game.

Break the Streak EE 0 : 15 CS

Break the Streak

The CS department has owned this court for years. Score more than 15 points in 40 seconds to end their streak.

Drag back from the ball, aim, and let go. Shots from behind the line are worth 3.

Based on a real result: EE beat CS for the department title in two straight seasons.

The pivot

2015 to 2017

In basketball, a pivot is how you change direction without giving up the ball. After two years of barely passing, I pivoted.

My degree was Electrical and Telecommunication Engineering, but my attention kept drifting to robotics. Dr. Omer Ahmed noticed, and opened that world up to me.

Assist: Dr. Omer Ahmed

Zaveen +2

My first robot was an unmanned ground vehicle for search and rescue. It mapped unfamiliar spaces with SLAM using a Kinect depth sensor, and it could be driven remotely while streaming live video from a long way off.

Zaveen +3

It won at FIT, Frontiers of Information Technology, in 2016.

Zaveen +3

Then it won again at the university's CEPEX Open House in 2017.

The UGV. Photo: SB Films
Demo day
Taking the award
SLAM Rescue Survivors 0/4

SLAM Rescue

A building has collapsed and nobody knows the layout. Drive the UGV in, let the depth sensor map it as you go, and find all four survivors before the battery runs out.

Arrow keys or WASD to drive. On touch screens, drag anywhere to steer.

Based on my first robot: SLAM mapping from a Kinect, remote driving and a live video feed.

The comeback

2017 to 2019

Zaveen +3

In 2017 I walked into NUST, the university I'd missed by a hair five years earlier, for a Master's in Robotics & Intelligent Machine Engineering.

Turnover

At the RISE Lab I worked on the lab's robots and built the application for a smart guidance robot that could lead visitors anywhere they needed to go. That project stalled before it shipped.

So I followed a new pull: machine learning, and a second mentor, Dr. Hasan Sajid.

Assist: Dr. Hasan Sajid

Zaveen +2

Meanwhile I took my first job, joining HyperNym as a graduate engineering trainee in IoT. I learned SoC firmware, designed PCBs in Altium and deployed hardware in the field. Our main product rode on petrol tankers and streamed their location and fuel level to a live dashboard.

LiveFuel
Tanker 07, on route

When the job started crowding out my master's, I stepped away.

Zaveen +3

Then Dr. Sajid sent me to Lahore, to a textile mill that needed automatic fabric defect detection. I did it end to end: mounted the cameras, collected and annotated the data, trained the model, quantized it to run in real time on an NVIDIA Jetson TX2, and built a GStreamer pipeline that handled multiple 4K streams live.

It worked, and the way it came together is what got me my next job.

Defect Hunter Caught 0

Defect Hunter

You're the model now. Fabric streams past the camera. Tap every hole, stain, broken thread and knot before it leaves the frame. Tapping clean fabric counts as a false alarm.

Click or tap defects. You'll get your precision, recall and F1 at the end.

Based on the Lahore deployment: cameras over the line, a quantized model on a Jetson TX2, every frame, all shift.

The league

2019 to now

2019 to 2024

siParadigm Diagnostic Informatics

AI Developer, then Team Lead, then Senior AI Developer. Remote, for a US diagnostics lab.

Zaveen +3

Five seasons in medical AI. I led Morph-AI for white blood cell detection and classification, OCR-AI for digitizing handwritten and printed lab forms, and FISH-AI for automated FISH testing. Together they cut pathology turnaround time and a lot of manual work.

Zaveen +2

As team lead I built a production video-conferencing platform with live call analytics, and a real-time humanoid animation engine in Three.js and FastAPI that turned plain-language prompts into joint trajectories, streamed over WebSockets.

Zaveen +2

In my last season I helped architect an AI-powered lab information system, designed for healthcare-grade availability and data isolation.

Off the clock, I got genuinely good at table tennis.

Break Room Rally You 0 : 0 Zaveen

Break Room Rally

First to 7. You're at the bottom, I'm at the top. Swipe across the ball as you hit it to add spin.

Move with your mouse or finger, or use the arrow keys.

The bot plays a lot like me. Mostly.
2022 to 2024

Voyagersol

Co-founder and CTO (consulting), alongside siParadigm.

Zaveen +2

A multi-vertical group spanning technology, education, marketing and entrepreneurship. I owned the technical strategy, architected cloud backends and infrastructure as code for V-Tech, and shaped products for V-Skill (e-learning, localized with Skillsoft), V-Launch and V-Comm.

2024 to 2025

AILA

Director AI. Artificial Intelligence Learning Assistance.

Zaveen +3

I led AI strategy and R&D for a multi-agent learning platform: personalized learning for students, lesson planning for teachers and term planning for school leaders. Under it sat a multi-tenant architecture of LLMs and retrieval, plus the standards, reviews, CI/CD and observability that let us ship new AI features safely.

2025 to now

Compact Machines

Head of AI at a Canadian sovereign-AI infrastructure company.

Zaveen +3

I'm architecting an AI due-diligence platform for M&A. A smart dataroom classifies and tags every document and checks it against advisor checklists, backed by a vector database and a knowledge graph. An analytics engine and CIM builder turn it into investor-ready material.

I also trained a domain-specific small language model for finance and M&A with LoRA, and built a LangGraph assistant whose answers stay grounded in source documents and respect who's allowed to see what.

Dataroom Dash Filed 0

Dataroom Dash

A deal closes in 45 seconds. Documents are pouring in. File each one in the right folder so the advisors' checklist fills up. At the buzzer, the platform flags whatever is still missing.

Click a folder, or press keys 1 to 6.

Based on what I'm building now: classification, checklist mapping and gap detection for due diligence.

Still hooping.

I never stopped playing basketball. Not at club level anymore, but I still show up to run.

Tap the card to flip it

Scouting report

Takes AI from a whiteboard to production and keeps it there. Comfortable in a room with founders, lab directors and engineers. Leads teams and still ships code.

GenAI and LLM platforms
Multi-agent systems, RAG, LangChain, LangGraph, Ollama, Groq, LoRA fine-tuning of small language models, vector databases, knowledge graphs
Computer vision and edge
PyTorch, TensorFlow, NVIDIA DeepStream, GStreamer, Jetson deployment, quantization, real-time 4K pipelines
MLOps and cloud
AWS, Docker, Kubernetes, CI/CD, observability, secure multi-tenant SaaS
Backend and product
Python, FastAPI, Django REST, NestJS, TypeScript, React, React Native, C++, WebSockets, Twilio
Robotics and IoT
SLAM, PCB design in Altium, embedded firmware, fault-tolerant field systems
Certifications
AWS Cloud Architect Associate (Udemy), Object Oriented Design and Design Patterns (Coursera)

Next possession

I'm based in Islamabad and work with teams in North America. If you're building AI that has to hold up in the real world, let's talk.

zaveenwaqar1993@gmail.com

Final Zaveen 39 07 The odds