Faster pathology pipeline
From preprocessing to review
AI software engineer · Applied AI researcher
I build AI systems that make it out of the notebook.
From pathology pipelines and aerial grounding to GPU-backed ML infrastructure and full-stack tools, I turn research-heavy problems into fast, reliable software people can actually use.
Selected organizations & deployments
Experience spanning applied research, public-sector education, and ML infrastructure, including product engineering inside a Y Combinator-backed company.Proof in production
Software that holds up beyond the demo.
I work across the full path—from model behavior to the infrastructure and interfaces that make it useful.
Faster deployments
Through a rebuilt CI/CD system
Government schools
Reached by a deployed LMS
Saved every month
By automating recurring company tasks
Experience
Where I’ve made an impact.
Production work spanning medical imaging, ML platforms, cloud infrastructure, and AI-enabled products.
01
Building AI-assisted pathology workflows for cell segmentation and HER2 / Chromosome 17 biomarker analysis.
- Rebuilt preprocessing, inference, post-processing, and review flows—cutting end-to-end latency by more than 20×.
- Created Vue canvas review tools, real-time WebSocket feedback, and reusable annotation operations.
- Set up AWS infrastructure and automated deployments for fast, iterative medical-image analysis.
PythonComputer VisionVueWebSocketsAWS
02
Engineering cloud-native foundations for repeatable ML training, deployment, and monitoring.
- Built Kubernetes-native ML infrastructure and GPU-backed model deployment workflows.
- Developed Pulumi infrastructure, Helm charts, CI/CD pipelines, and canary releases on AKS.
- Modeled workload telemetry to forecast capacity and guide operational decisions.
KubernetesKubeflowPulumiHelmAzure
03
Shipped AI-enabled products across manufacturing, retail, energy analytics, ERP, and IoT.
- Integrated computer-vision workflows for optical-fiber alignment and smart-cart checkout.
- Reduced deployment time by 66% with a redesigned CI/CD system and parallelized tests.
- Built zero-downtime Kubernetes delivery and automation with Ansible, Prometheus, and Grafana.
Node.jsReactKubernetesAnsibleGrafana
Selected work
Built to solve, not just to showcase.
Four shipped systems spanning clinical AI, developer tooling, computer vision, and education infrastructure.
01Clinical AI · Flagship
Biomarker Quantification AI
An AI-assisted pathology platform for HER2 / CEP17 dual-probe analysis—from cell segmentation and signal counting to ratio calculation, expert review, and concordance analysis.
01 Automated cell detection and biomarker counting
02 HER2 / CEP17 ratios across 25-cell clusters
03 Human-in-the-loop correction and review
02Featured build
Open-source developer tooling
ML Canvas
A reusable Vue canvas library for ML annotation workflows with rectangle, polygon, freeform, inspect, click, and delete modes—plus dual display and source-image coordinates.VueCanvas APIJavaScriptnpm
03Featured build
Classical computer vision
Fingerprint Extraction
A command-line computer-vision pipeline that detects four fingers from a hand image and extracts each unique fingerprint for inspection.PythonOpenCVNumPyImage Processing
04Featured build
Deployed education platform
Government School LMS
Created and deployed a learning management system across 30 government schools in Pakistan, supporting real classroom operations at scale.Full-stackLMSDeploymentPakistan
Also shipped
More product surfaces.
A broader selection of client and product environments connected to my engineering work.
01AutomaSmart AC control & energy analytics02Cowlar StorybookVue 3 component library03Cowlar ShopOnline grocery delivery04Restaurant OSRestaurant operations & POS05ThanosML annotation & deployment06Canary DeploymentControlled rollout & monitoring07RetailOSRetail operations & analytics08Bijli MetricsEnergy monitoring & optimization09Access ControlAccess & attendance management10Human D/CodedAI meal planning & nutrition
Capabilities
Research → systems → product.
The best AI work needs more than a model. I connect each layer into one dependable system.
01Understand
→Frame the real problem.
02Engineer
→Build the right system.
03Ship
Measure it in use.
Applied AI
Models grounded in real inputs, measurable behavior, and usable outputs.
- Python
- PyTorch
- TensorFlow
- Computer Vision
- Medical Imaging
- Visual Grounding
- Transformers
- Diffusion Models
- RAG
- OpenCV
ML Systems
Infrastructure that makes training and serving repeatable, observable, and safe.
- Kubernetes
- Kubeflow
- Docker
- MLflow
- Helm
- Pulumi
- CI/CD
- AWS · Azure
Product Engineering
Interfaces and APIs that turn technical capability into a coherent product.
- Next.js
- React
- Vue
- Node.js
- REST APIs
- WebSockets
- SQL
- MongoDB
Research
Research shaped by real deployment constraints.
My research sits at the intersection of multimodal understanding and efficient generative modeling. I study how models can reason across language and visual data while improving inference efficiency and task performance.
Efficient generative modeling
Diffusion-based language models.
Actively researching the speed–quality trade-off in iterative language generation. The goal is to reduce sampling cost and latency while preserving—or improving—task performance.UAV visual grounding
Language-guided target localization.
Worked on grounding tasks in UAV imagery, connecting natural-language descriptions to spatial predictions in complex aerial scenes.Medical AI foundation
Previous work in medical image understanding.
My earlier research spans dental, chest X-ray, breast ultrasound, retinal imaging, and production pathology tooling. It includes a peer-reviewed paper and public implementations across segmentation and attention-based classification.
01Publication · Co-first authorMask-transformer-based networks for teeth segmentation in panoramic radiographsBioengineering · Transformer segmentation02Production research platformBiomarker Quantification AICell segmentation · HER2 / CEP17 analysis03Research repositoryEnhanced Breast Tumor SegmentationTransUNet · Medical segmentation04Research repositoryLung Segmentation in X-ray ImagesU-Net · ResNet · Semantic attention05Research repositoryChest X-ray ClassificationXception · CBAM attention06Research repositoryFundus SegmentationCustom U-Net · DRIVE dataset
Recognition
Selected recognition.
Employees of 2023 · Cowlar Design Studio
NaSCon Data Quest · 2023
Start a conversation
Have a difficult AI problem?
I'm always interested in ambitious engineering work at the intersection of AI, infrastructure, and product.
mutti.rehman1122@gmail.com↗