AI / ML Engineer

HamzaBasharat

AI that survives production — not notebooks.

Lahore, PakistanOpen to roles & freelance

I build computer-vision pipelines, RAG agents and the automation around them — from raw dataset to a deployment you can monitor and hand off. Shipped into manufacturing QC, industrial safety, UAV perception and enterprise tooling.

Delivered
MagicQC — web + desktop, in production
Computer vision, end to end
90%+
Measurement accuracy
MagicQC
0.81
mAP50, fabric defect
YOLO + PatchCore
1st
National podium finishes ×2
IEEE Hackathon · ICAT Robotics
PyTorchYOLOv8 / v11OpenCVPatchCoreONNXOAK-1W / DepthAITensorRTFastAPIDockerAWS EC2LangGraphClaude APIFAISSVision TransformersPaddleOCRGazebo / PX4React / Next.jsMLflow

Selected work — shipped & in the lab

Shipped products, measurable outcomes

The measurement view: a shirt checked to spec, PASS, every panel dimension in centimetres.
Computer Vision500+Pieces / day

MagicQC

2024-2026

Finished garments are size-checked by hand with a tape measure, so QC is slow, inconsistent between operators, and impossible to audit after the fact — and the brands, operators and purchase orders behind that QC line have no system of record at all.

  • PyTorch
  • OpenCV
  • FastAPI
  • Electron.js
Live overlay: players tracked by role (coach / student), ball speed, rally and shot counts, wall-target distance.
Computer Vision22.8–76.1Ball speed range

Most tennis computer-vision systems assume a fixed camera, a regulation court, labelled training data and a GPU. Real coaching footage — a handheld camera, a driveway wall, no dataset — has none of those, so the tooling that exists does not run on it.

  • YOLO11x
  • YOLO11x-pose
  • Ultralytics
  • BoT-SORT
One frame, four stages: raw cloth, PatchCore anomaly heatmap, YOLO detections, fused result.
Computer Vision0.81mAP50

Fabric defects are caught by human inspectors watching moving cloth, so rare and unfamiliar defect types slip through and nothing is logged for the mill to act on.

  • YOLO
  • PatchCore
  • OAK-1W / DepthAI
  • PyTorch
Gazebo simulation of the UAV airframe, used to validate the perception stack before flight.
Computer Vision

In a disaster zone, finding people fast is the whole problem, and a human pilot watching a UAV feed cannot search a wide area quickly or reliably.

  • Python
  • OpenCV
  • Gazebo
  • PX4 SITL
Operations view: the live dock feed with people and vehicles detected, beside the turnaround event feed.
Computer Vision1st placePlacement

A loading dock’s turnaround - gate, docked, unloading, gone - is tracked by eye and on clipboards, so a slow stage is only noticed after the truck has left and the detention charge has already landed.

  • Python
  • OpenCV
  • Object detection
  • Video analytics
The CV Recruiter ranking view: a job description on the left, candidates ranked by match score on the right, each score citing the exact line in the CV that earned it.LLM / RAG

Screening a stack of CVs against one job description is slow, inconsistent, and the reasoning behind a rejection is never written down.

  • LangGraph
  • Claude API
  • FAISS
  • FastAPI

In the lab

YOLOv8 locating a vehicle number plate and PaddleOCR reading the characters off it in real time.OCR
Licence-plate OCR
DeepLabV3 segmenting a walking person from the background frame by frame.Segmentation
Person segmentation
Twenty-one hand landmarks tracked across a moving hand.Tracking
Hand keypoint tracking
Faces detected and blurred automatically as people move through the frame.Detection
Face anonymisation
Head pose and gaze direction estimated from a webcam feed and drawn as a vector.Pose
Gaze estimation
Full-body skeletal keypoints — torso, limbs, face — tracked frame to frame as a person moves.Pose
Full-body pose tracking
Woven fabric beside its PatchCore anomaly heatmap, defects lighting up as the cloth moves.Anomaly
Fabric anomaly heatmap
Chocolates tracked and tallied in and out of frame as they move along a conveyor.Tracking
Conveyor counting
Lane lines extracted from a LiDAR point cloud with threshold and region-of-interest filtering.Depth
3D LiDAR lane lines
Twenty-eight people detected in one frame with 42 unique tracking IDs held across the scene.Tracking
Crowd counting + re-ID
Detectron2 instance masks over a car's dented and cracked panels for insurance assessment.Segmentation
Vehicle damage segmentation
A YOLO detector separating an Asian hornet from a native bee on a monitoring plate.Detection
Invasive-hornet detection
A face detected and labelled with an estimated age and gender from a single webcam frame.Detection
Age & gender estimation
A face classified as happy in real time with the detection box drawn around it.Detection
Emotion recognition
A vehicle detected at an automatic aluminium boom barrier to trigger the gate.Detection
Barrier-gate vehicle trigger
An n8n workflow pulling a note from Notion, formatting it, and posting to LinkedIn on a schedule.Automation
Content automation (n8n)

Proof, not promises

3
National podium finishes
IEEE · ICAT · NAIS
13
Certifications
AWS ML · Microsoft · DeepLearning.AI
1
Industrial CV workshop taught
PTUT Lahore · with Robionix
3
Client geographies
US · Canada · Pakistan

Awards

  • 1stDock Vision AIIEEE Hackathon · 2024
  • 1stICAT National Robotics CompetitionICAT
  • Runner-upNational AI Sprint (NAIS)National Centre of Physics · 2025
  • ExhibitorMagicQC at TextileAsia32nd Edition · Lahore Expo Centre · NUTECH

Teaching

Hamza presenting from the lectern to attendees at rows of workstations in the PTUT computer lab.

CPD Workshop — Applications of Vision AI in Manufacturing Industries

Punjab Tianjin University of Technology (PTUT), Lahore · conducted by NUTECH in collaboration with Robionix Technologies

PTUT Township Campus, Lahore · 2-3 July 2026

Instructor — Robionix engineering team, with Prof. Dr. Awais Yasin (Founder & CEO, Robionix)

  • Day 1: computer vision and AI fundamentals, YOLO detection and classification, Python/Django web APIs, MySQL from shop-floor to manager PC, React/Next.js dashboards, real-time industrial camera interfacing.
  • Day 2: a full fabric-defect-detection build — dataset preparation and training, back-end API configuration, real-time ERP quality dashboards, the automated textile pipeline demo, and a prompt-engineering session.

From the field

Hamza on the MagicQC stand at TextileAsia, Lahore Expo Centre.
TextileAsia · Lahore
The Robionix team on the MagicQC stand at TextileAsia, Lahore Expo Centre.
The team · TextileAsia
Hamza demonstrating the MagicQC size-measurement rig to a visitor at the TextileAsia stand.
MagicQC demo · TextileAsia
The AI-enabled fabric-defect inspection rig on the mill floor — camera and lighting over the fabric roll, operator PC alongside.
Defect-detection rig · on the floor
The autonomous sorting robot Hamza’s team built for the ICAT National Robotics Competition, bins labelled metal / plastic / unknown.
ICAT robotics · 1st
Hamza in a working discussion with a client at the exhibition table.
Client meeting · Lahore

What I do

The stack I build and maintain on

  • YOLOv5/v8/v11/26
  • OpenCV
  • PatchCore
  • Vision Transformers (ViT)
  • Vision-Language Models
  • PaddleOCR
  • Tesseract
  • Tracking
  • Pose estimation
  • Segmentation
  • Anomaly detection
  • PyTorch
  • TensorFlow
  • Keras
  • Hugging Face Transformers
  • Diffusers
  • Diffusion models
  • Transfer learning
  • Fine-tuning
  • Data augmentation
  • Anthropic Claude API
  • OpenAI API
  • LangChain
  • LangGraph
  • RAG pipelines
  • FAISS
  • Chroma
  • Prompt engineering
  • Context management
  • OAK-1W / DepthAI
  • ONNX
  • Embedded inference
  • Docker
  • AWS EC2
  • Azure AI
  • REST API design
  • Model versioning
  • Monitoring
  • CI/CD
  • Python
  • C++
  • C
  • JavaScript
  • MATLAB
  • SQL
  • FastAPI
  • Flask
  • Streamlit
  • React
  • Next.js
  • Node.js
  • Laravel
  • MySQL
  • MongoDB
  • Roboflow
  • CVAT
  • Gazebo
  • Linux

AI / ML Engineer · Lahore, Pakistan · Open to relocation

A model that stays in a notebook is worth nothing to you. I build the ones that reach production — and keep running after the hand-off.

I build computer-vision pipelines, RAG agents and the automation around them — from raw dataset to a deployment you can monitor and hand off. Shipped into manufacturing QC, industrial safety, UAV perception and enterprise tooling.

Experience

AI & Computer Vision Engineer

Aug 2024Jul 2026

Robionix Technologies · Islamabad, Pakistan

  • Led MagicQC end to end — an automated size-measurement system for apparel production, live on AWS EC2 at 500+ items/day and 90%+ accuracy. Showcased at My Karachi Expo and TextileAsia, Lahore. Case study ↗
  • Shipped a detection and pose-estimation suite hitting 92%+ mAP at 15+ FPS on constrained edge hardware. Case study ↗
  • Built enterprise RAG and conversational agents served as REST APIs into Laravel, React and MySQL production backends. Case study ↗
  • Automated tag, label and document OCR feeding downstream QC reporting and inventory records.
  • Used diffusion-based augmentation and ViT/VLM fine-tuning to close class gaps in scarce industrial datasets.

AI Engineer (Contract)

Jul 2025Sep 2025

Essenceware Technologies · Pakistan

  • Trained, deployed and handed off real-time PPE and workplace safety detection with inference dashboards and model versioning. Case study ↗
  • Integrated a real-time pose-estimation and activity-monitoring module into the Essenceware product stack.

AI & UAV Engineer (Intern)

Jun 2025Sep 2025

NESCOM, National Development Complex · Islamabad, Pakistan

  • Built detection and tracking pipelines and Gazebo simulation for SkyResQ, an autonomous UAV disaster-response perception system. Case study ↗
  • Presented results directly to Pakistani government stakeholders evaluating the system for national emergency preparedness.

Skills

Computer Vision & Video Analytics
YOLOv5/v8/v11/26 · OpenCV · PatchCore · Vision Transformers (ViT) · Vision-Language Models · PaddleOCR · Tesseract · Tracking · Pose estimation · Segmentation · Anomaly detection
Deep Learning & Generative AI
PyTorch · TensorFlow · Keras · Hugging Face Transformers · Diffusers · Diffusion models · Transfer learning · Fine-tuning · Data augmentation
LLM & Agentic AI
Anthropic Claude API · OpenAI API · LangChain · LangGraph · RAG pipelines · FAISS · Chroma · Prompt engineering · Context management
Edge AI & MLOps
OAK-1W / DepthAI · ONNX · Embedded inference · Docker · AWS EC2 · Azure AI · REST API design · Model versioning · Monitoring · CI/CD
Backend, Tooling & Languages
Python · C++ · C · JavaScript · MATLAB · SQL · FastAPI · Flask · Streamlit · React · Next.js · Node.js · Laravel · MySQL · MongoDB · Roboflow · CVAT · Gazebo · Linux

Awards

  • 1st Dock Vision AIIEEE Hackathon (2024)
  • 1st ICAT National Robotics CompetitionICAT
  • Runner-up National AI Sprint (NAIS)National Centre of Physics (2025)
  • Exhibitor MagicQC at TextileAsia32nd Edition · Lahore Expo Centre · NUTECH

Certifications

  • AWS Certified Machine Learning Engineer - AssociateAmazon Web Services
  • AI & ML Engineering Professional CertificateMicrosoft
  • Advanced Computer Vision with TensorFlowDeepLearning.AI
  • Exploratory Data Analysis for Machine LearningIBM
  • Azure AI / Computer VisionMicrosoft
  • LangChain for LLM Application DevelopmentDeepLearning.AI
  • Introduction to Embedded Machine LearningEdge Impulse
  • Claude 101 & AI Fluency Framework FoundationsAnthropic
  • First Principles of Computer VisionUniversity of Colorado Boulder
  • Python for EverybodyUniversity of Michigan
  • Edge AIEdge Impulse
  • AI for All: From Practice to Gen AINVIDIA
  • Prompt EngineeringGoogle
BS Computer Engineering, National University of Technology (NUTECH) (2022 - 2026)Full CV →Download PDF ↗

Frequently asked

How do engagements usually start?+

A feasibility read on a sample of your own data, at no cost — you get a straight yes or no on whether it will work. Then a scoped, fixed-price pilot of two to three weeks. A full build only after the pilot has proven the approach on your data, not a demo set.

What does a project cost?+

Pilots start at $1,500. Full deployments run $5,000–15,000 depending on scope. Priced per project, not hourly — the number is fixed the moment we agree it, and it covers the hand-off, not just the model.

You are not in our timezone — how does that work?+

Calls happen in your hours. Work ships in weekly increments, so you see progress instead of a reveal at the end. The pilot exists precisely so you can test how we work together on something small before committing to a build.

Do you work alongside an in-house team?+

Usually, yes. Most of this is the specialist vision or retrieval piece a team does not want to pause its roadmap to build — I take that part and hand back clean, documented interfaces.

What can you not do?+

I am a vision, ML and retrieval specialist — not a full data-engineering or DevOps team. I build the model and the inference service and hand off clean interfaces; I do not run your data warehouse or own your cloud estate. Where a project needs that, I say so up front and help you scope it out.

What people say

  • Hamza works both sides of computer vision, research and production, which most people don't. He's built and deployed real-time detection systems running on edge hardware in live industrial settings, not just in a notebook. He also led an AI-powered measurement platform from prototype through to a live cloud deployment on his own. On top of that, he's comfortable building with modern LLM and retrieval systems too. That range is rare enough to be worth a conversation if you're hiring for computer vision or applied AI.

    Nasir Mehmood

    Head of Retail Sales · Security General Insurance Company Ltd.

    LinkedIn ↗ — recommendation from Nasir Mehmood
  • I worked with Hamza on a few AI and backend projects during and after university. He’s solid with YOLO-based computer vision work and comfortable building out the backend to actually deploy it, not just get a model working in a notebook. He asks good questions, follows through, and I never had to double-check his work. Good person to have on a project.

    Sami Uddin

    AI / Computer Vision Engineer · Codexia Tech

    LinkedIn ↗ — recommendation from Sami Uddin
  • I highly recommend Hamza for his exceptional work as a Full Stack AI Developer. He demonstrated strong technical knowledge, clear communication, and consistent problem-solving ability throughout our collaboration. He successfully developed and integrated LLM and RAG-based chatbot solutions that were practical, scalable, and built around real user needs. His ability to handle both frontend and backend while delivering intelligent AI features reflects a rare combination of technical depth and product understanding. Hamza is dependable, quick to adapt, and committed to delivering quality results. He is a developer I would confidently engage again on any AI-driven project.

    Saira Gillani

    Reliability Analysis · Statistical Modeling · Computer Vision · AI · Client

    LinkedIn ↗ — recommendation from Saira Gillani
  • I highly recommend Hamza for his exceptional work in developing the platform using AI technologies. Hamza demonstrated strong innovation, technical capability, and problem-solving skills throughout the development process. He successfully leveraged artificial intelligence to create a practical and forward-thinking solution focused on safety, monitoring, and operational efficiency. His ability to combine AI concepts with real-world applications shows both technical expertise and a strong understanding of user needs. Hamza is highly motivated, quick to learn new technologies, and committed to delivering quality results.

    Mohammed Faisal Ghayas

    Senior Account Manager — Building Automation (BAS/BMS) · Client

    LinkedIn ↗ — recommendation from Mohammed Faisal Ghayas
  • I had the opportunity to work with Hamza in the AI automation and computer vision space, and I’ve been consistently impressed by his technical skills and problem-solving mindset. He has a strong understanding of AI development, LLM pipelines, and computer vision systems, and he approaches projects with both creativity and attention to detail. Hamza is reliable, quick to learn, and always focused on building practical solutions that create real value. I’d highly recommend him to anyone looking for a skilled and dedicated AI developer.

    Haris Ai

    Multi-agent AI systems · OpenClaw + n8n · Independent

    LinkedIn ↗ — recommendation from Haris Ai

Let's build something that ships.

Open to full-time roles, contract engagements and freelance work — with clients in the US, Canada and Pakistan.

hamzabasharat2004@gmail.comEmail me

+92 300 6547302LinkedIn ↗

Lahore, Pakistan · Open to relocation