Haris Rehman

Technical Product Manager | Ex-VyroAI APM | AI & Big Data Specialist

Based in: Alberta, CanadaCurrently: AI Researcher, University of AlbertaExperience: 2 years

Technical Product Manager with a background in scaling AI products and deep technical research. Previously an Associate Product Manager at VyroAI, where I drove 6x growth in MAUs and 10x growth in AI Chat users. Currently an AI Researcher at the University of Alberta, bridging the gap between complex multimodal modeling and product-led growth. Passionate about building high-impact tools for developers and early-stage startups.

Experience

AI Researcher

University of AlbertaEdmonton, CanadaworkJun 2026 – Present

Summary

Developing multimodal generative and foundational models using the MIMIC-IV dataset, integrating 12-lead ECG waveforms with longitudinal EHR data for population-scale cardiovascular modeling.

What I did

Developing multimodal generative and foundational models integrating 12-lead ECG waveforms with longitudinal EHR data (diagnoses, procedures, labs, medications, hospitalization timelines) for population-scale cardiovascular modeling.
Building large-scale patient embedding models that learn temporal disease trajectories from aligned ECG and administrative health record sequences.
Forecasting clinical outcomes including rehospitalization, MACE, & mortality using multimodal time-series inputs.

Generative ModelsFoundational ModelsECG WaveformsEHR DataTime-series

Associate Product Manager

VyroAI - ChatlyIslamabad, PakistanworkOct 2025 – May 2026

growth in AI Chat users: 10xgrowth in AI Chat users
50% uplift in DAUs: 3550% uplift in DAUs
improvement in release velocity: 20%improvement in release velocity

Summary

Scaled overall product from 200K to 1.2M+ MAUs (6x growth) by driving feature prioritization, growth experiments, and cross-functional execution.

What I did

Grew AI Chat from 7K to 70K+ users (10x) and launched AI Docs from 0 to 2K+ users within 30 days, achieving rapid product-market validation.
Increased user engagement with 35-50% uplift in DAUs through onboarding improvements, retention strategies, and iterative UX enhancements.
Contributed to monetization efforts, supporting revenue growth to the $6.5M range via pricing experiments, funnel optimization, and Stripe billing improvements.
Drove data-informed decisions by implementing end-to-end event tracking (GTM → Mixpanel), building real-time dashboards, and automating feedback reporting across teams.
Collaborated cross-functionally with Engineering, Design, SEO, Branding, Support, and QA to identify pain points, prioritize roadmap, and improve release velocity by 20%.
Scaled team operations from 5 to 35+ members, improving cross-functional coordination, delivery efficiency, and execution bandwidth.
Built 5+ internal tools to automate workflows, significantly reducing manual effort and improving team productivity.
Improved platform reliability by contributing to QA automation workflows, reducing production issues by 50% and increasing release confidence.
Synthesized 3,000+ user feedback points/month into actionable insights to inform product strategy and feature development.

Results

6x growth in MAUs (200K to 1.2M+)
Revenue growth to $6.5M range
50% reduction in production issues

GTMMixpanelStripeQA AutomationDashboards

Machine Learning Engineer

Cowlar Design Studio (YC W-17)Islamabad, PakistanworkApr 2024 – Apr 2025

production capacity boost: 40xproduction capacity boost
accuracy in fiber sorting: 96%accuracy in fiber sorting
increase in AWS cable production: 500%increase in AWS cable production

Summary

Assisted with automation efforts for a factory in North America, boosting production capacity by 40x for a key Amazon-partnered Fiber Optics supplier. Developed a system leveraging ML and CV to align fibers with 5-micrometer precision, scaling monthly output from 240 to 9,600 cables.

What I did

Enhanced the Fiber Organizer Robot, integrating ML and CV to sort 250-micron fibers with 96% accuracy, increasing AWS cable production by 500% (from 1 to 6 cables per hour).
Pushed AI solutions to production by setting up CI/CD pipelines with Docker whilst load balancing deployment servers using techniques like Nginx.
Developed and deployed real-time AI solutions for customer-centric applications, including an Action Recognition system for Smart Carts achieving 95% accuracy and an Azure OpenAI chatbot to revolutionize online retail customer experiences.
Optimized model inference speeds using NVIDIA-enabled dockerization, code optimization, and dimensionality reduction, reducing inference times from 200ms to less than 100ms per request.
Used Docker Swarm orchestration to set up distributed GPU compute across nodes for training models 3-4x faster.

Results

95% accuracy in Action Recognition
Reduced inference time from 200ms to <100ms
3-4x faster model training

MLCVDockerNginxAzure OpenAINVIDIADocker SwarmGPU ComputeHadoop HDFSApache SparkHive

Skills

Technical

Computer Vision

OpenCV

YOLO

Deep Learning

TensorFlow

PyTorch

Generative AI

Action Recognition

Data Science

MLOPS

Scikit-Learn

NumPy

Pandas

Hugging Face Transformers

Docker

FASTAPI

Edge Model Deployment

GPU Optimization

Distributed Training

Git

MediaPipe

TorchServe

PostgreSQL

Natural Language Processing (NLP)

Remote Docker Image Registry Management

Postman

Conda

poetry

Matplotlib

MySQL

Azure AI Studio

Azure Machine Learning Studio

Kubernetes (Docker Swarm)

CI/CD

MLFlow

TensorBoard

TensorFlow Serving

Uvicorn

Flask

Nginx Load Balancing

Redis

MongoDB

StatsModels

JSX

Reinforcement Learning

Collaborative Filtering

React

RedHat tmux

Hadoop HDFS

Apache Spark

Hive

MQTT

Prometheus

AlertManager

Insomnia

Zookeeper

Areas of expertise

Team Collaboration

Task Management

Critical Thinking

Project Management

Problem-Solving

Languages

Urdu

English

Punjabi

German

Japanese

Projects

Fiber Organizer & Termination Line Robot

Contributorprofessional

Contributed to the development of the patented (US 12001073 B2) machine, currently deployed in Aurora, IL, streamlining operations and reducing manual intervention. Contributed to the automation of a factory for an Amazon-partnered Fiber Optics supplier, increasing production capacity by 40x (from 240 to 9,600 cables per month). Assisted the development of the Fiber Organizer Robot, a patented system utilizing CV and ML to sort 250-micron fibers, increasing production capacity by 500% and reducing sorting time from 3 minutes to 8 seconds. Integrated multiple fiber termination processes into an automated system, achieving 5-micrometer precision and a 96% success rate, producing 50 AWS cables per day.

MLCVAutomation

Smart Shopping Cart

Contributorprofessional

Contributed to the development and improvement of Cowlar Design Studio's smartcarts for Al-Meera Supermarkets in Qatar. Optimized and improved the Action Recognition architecture for predicting customer interactions with smartcarts, achieving 95% accuracy in real-time on edge devices. Continuously cleaned and improved the data pipeline through rigorous validation and testing to debug and better the model performance.

TensorflowMoViNetMQTTMLflowNGINXDocker

B.Sc. in Software Engineering

National University of Science and Technology (NUST)Software Engineering

Recognition

Awards

Cambridge Outstanding Learner Award

Cambridge Assessment International Education

Ranked 2nd in 5 different subjects in A Levels among all the students of Punjab district.

National Science Talent Contest (NSTC)

NSTC Pakistan

Stood in the top 50 selected candidates in mathematics from all over the country who then competed for a chance to represent the country in the international math Olympiad.

Certifications

Patents & publications

Fiber Organizer Robot System

patentUS 12001073 B2

Patented system utilizing CV and ML to sort 250-micron fibers.

Social impact & activities

Open Source Contributor

Ultralyticsother

Contributed open source to Ultralytics. Enabled distributed parallel computing for on-premises architecture for YOLO training. Connected the dots and got a new version released with the new support available for everyone.

What drives the work

Motivations: Technical Product Management
Product Growth
Early-Stage Startups

Hire HarisReach out about a role, a contract or a conversation.For recruiters

Haris's twin is AI, it can make mistakes.

More people on Kanvis

Browse more people Create your own Kanvis