ABOUT
My work begins with understanding the problem, not choosing the technology.
At Nolyth, I lead the development of intelligent products that combine Generative AI, computer vision, IoT, automation, voice technology, biometrics, and scalable cloud platforms.
Depending on what the product requires, the solution might involve an AI agent, a computer vision pipeline, connected IoT devices, a recommendation engine, an automated workflow, or a complete cloud-based platform combining several of these capabilities.
I build AI products for healthcare, retail, fintech and ESG operations: computer vision, voice agents, biometric verification and IoT platforms, taken from discovery to production for clients in Pakistan, the Gulf, the UK and the US.
NOW Co-Founder, Head of Product & Technology · Nolyth — CTO · TripVlog ·
Product discovery and use-case validation
Technical architecture and system design
AI and machine learning development
Backend, cloud, and platform engineering
Data pipelines and third-party integrations
Product delivery, deployment, and scaling
Engineering leadership and team building
SELECTED WORK
Production systems, measured by what changed for the operation.
Cashierless Retail
A retail operator wanted checkout removed entirely. Earlier vision-only pilots failed on occlusion, crowded shelves and look-alike products, and every mistake was a refund and a lost shopper.
Multi-camera person tracking with cross-camera ReID, shelf event detection, and weight-sensor fusion arbitrating between them. GStreamer pipelines on CUDA, MLflow for the model lifecycle, sub-second decisions.
A working grab-and-go store for enterprise clients. Basket accuracy high enough to run unattended, and a system that tells operators when it is unsure instead of guessing.
Fleck
Banks and telcos needed fingerprint verification without a hardware scanner in every branch and shop. Phone cameras vary wildly, spoofing attempts are constant, and verification had to be near real time on consumer devices.
Contactless four-finger capture, GAN-based enhancement to scanner quality, anti-spoofing models, FastAPI services with a full MLOps loop.
Live at a tier-1 bank and a national telecom operator, processing 10,000+ verifications a month. Verification moved from the branch counter to the customer’s hand.
Althea AI
High-volume clinics lose staff hours to scheduling, verification and repeat questions on the phone. Callers are patients, so latency, tone, and knowing when to hand off matter more than raw capability.
An autonomous medical representative on a QLoRA-tuned Ultravox backbone: patient verification, appointment scheduling and inquiries in real time, live API integration, deployed on GCP with a clear boundary on what the agent never decides.
98% task completion across 300,000+ conversations, 600,000+ staff minutes saved, and routine calls handled around the clock without a queue.
N3XUS
ESG programmes start with the framework and work backwards to the evidence, so five reports about the same operations disagree. Sovereign funds and enterprises needed one evidence foundation across carbon, capital and compliance.
Sustainability intelligence combining AI, cryptographic proof and IoT sensor arrays, turning operational signals into decision-grade evidence that any framework can draw from.
2.5M+ tCO₂ and $850M+ in carbon value tracked with a verifiable trail, reporting produced from evidence instead of assembled around it.
WRITING
Where AI creates value, and where it only looks like it does.
The Biggest ESG Reporting Mistake Is Starting With the Report
The Most Valuable AI in Healthcare May Never Diagnose a Patient
Your Cameras Are Making Decisions. Can You Defend Them?
Why Vision-Only Warehouse Automation Breaks
We Are Optimizing Serendipity Out of the Internet
A Viral Hotel Video Can Still Produce Zero Bookings
Pakistan Should Copy India’s Sequence, Not Its Scale
The Most Elegant Equation in whole Math
Most ESG programmes begin with the framework: GHG Protocol, GRI, SROI, a regional MRV scheme. Then they work backwards to find the evidence. That order is why five reports about the same operations disagree with each other, and why the fifth costs as much as the first.
CONTINUE ON LINKEDIN →
Mean Shift Clustering (+implementation)
EinSum: Einstein Sum, Making Deep Learning Better
CUDA C/C++ on Google Colaboratory
The Oracle Problem & Decentralized Oracles
EXPERIENCE
From engineer to founder, staying hands-on the whole way.
National University of Computer and Emerging Sciences
HOW I WORK
Three ways to work with me. All of them start with the operation, not the model.
For founders and operators who already have a team and need a second brain on AI product decisions: architecture reviews, model and vendor choices, hiring, and the honest call on whether a feature should exist.
For companies with a real product problem and no technical leader yet. I own the roadmap, hire and lead the first engineers, and hand over to a full-time CTO when the company is ready for one.
For a defined product that needs to exist in production. I scope it, my team at Nolyth builds it, and we deploy it into your operation with the people who will run it. You own the code and the data.
Sixty minutes on what you are trying to change and what has been tried. I will tell you if I am the wrong person.
Two pages: the problem, what I would build or advise, what I would leave out, and what it costs. Yours to keep either way.
The first milestone is something working for one real user inside your operation. Not a demo.
We continue, change shape, or stop, based on what the pilot showed. Every engagement has a clean exit written in.
RECOMMENDATIONS
Zohaib was the only person in the room who could talk to our clinicians in the morning and ship the pipeline in the afternoon. He treated our operational mess as the product, not a distraction from it.
"He scoped our biometric onboarding down to what actually mattered for compliance, then delivered a system that has run in production without drama."
Khalid Al MansooriHead of Digital, financial services group · Dubai
"Rare combination: strong enough on the architecture to challenge our engineers, clear enough with the board to explain why it mattered."
James HollowayManaging Partner, venture studio · London
"Our computer vision deployment had failed twice before. Zohaib found the sensor fusion problem in week one and had it stable in a live store by week six."
Daniel ReyesDirector of Retail Technology · Austin, TX
"He built our voice agent to handle the calls our staff dreaded, and was honest about the ones it should never take. That judgement is why we trust him."
Dr. Priya RamanClinic Group Medical Director · Manchester, UK
"Zohaib joined as a technical advisor and left us with a working GTM engine, a data model, and a team that knew how to run it without him."
Fatima Al SayeghFounder, B2B SaaS · Riyadh
"Calm under pressure, direct about trade-offs, and generous with the team. The engineers who worked under him got better, quickly."
Michael OseiCTO, logistics platform · New York
"We came to him with a slide deck and left with a shipped platform. He pushed back on half the feature list and was right about all of it."
Omar HaddadChief Digital Officer, hospitality group · Abu Dhabi
"The most useful technical co-founder conversation I have had. He measured everything against whether a real customer would pay for it."
Charlotte BeaumontGeneral Partner, seed fund · Edinburgh
CONTACT
Building an AI product or exploring where AI creates value? Share what you are working on.
m.zohaibnasir6@gmail.com