Machine Learning Engineer | Computer Vision, Deep Learning & MLOps | PyTorch, FastAPI, Docker, Kubernetes | ACR 2026 & MICCAI 2026
Final-year BSc CS at FAST-NUCES Faisalabad (2026) · freelance engineer since 2023 building production ML systems and full-stack web products for local and international clients.
Open to AI/ML engineering roles, research collaborations, and PhD opportunities · Faisalabad, Pakistan (Remote)
- 🔭 Building production computer vision and ML features at HusQuay Solutions (via Crew Innovations)
- 🏥 Shipping DocDesk, a healthcare RAG service with a from-scratch LLM evaluation harness
- 🏆 Accepted at Learn2Reg 2026 (MICCAI), invited oral presentation
- 📄 Co-author, abstract accepted at ACR 2026 (American College of Radiology)
Property-Tech Vision AI Platform - flagship, confidential
A production computer vision and VLM pipeline that turns property photos into structured, customer-ready outputs for a property-tech SaaS platform serving US customers. Production details are confidential, happy to walk through the architecture in interviews.
FastAPI Computer Vision VLM MLOps
RadGuard · live demo
Catches 4 classes of error in AI-generated chest X-ray reports. BioViL-T + CXR-BERT cross-attention over 74K MIMIC-CXR samples, macro F1 0.66 after root-causing a patient-level data leakage bug.
PyTorch FastAPI Docker Kubernetes
PIT-Scorecard · live demo
Measured a +0.30 AUC leakage gap hiding in a standard credit scorecard, then shipped the honest version: a WOE/IV scorecard validated on a strict out-of-time split across 2.26M loans.
Python LightGBM FastAPI DuckDB
DocDesk
A medical RAG service that cites its sources, abstains when they do not cover a question, and refuses dose or diagnosis questions outright, validated by an in-house LLM evaluation harness with paired bootstrap intervals.
FastAPI PostgreSQL pgvector Docker
E2E-VarNet MRI Reconstruction · live demo
Reproduced E2E-VarNet for 4x-accelerated MRI reconstruction and added a checkpoint-ensemble uncertainty signal that flags out-of-distribution scans, backed by Evidently AI drift monitoring.
PyTorch Gradio Evidently AI
14 more projects
- Learn2Breath (Learn2Reg 2026) - Training-free lung CT registration pipeline, accepted at MICCAI 2026 with an invited oral presentation
- InferBench — Benchmarked PyTorch, ONNX Runtime, and TensorRT; a Redis cache cut repeat requests 2,377x
- ProgressGAN-CXR — Four severity-conditioned GANs simulating COVID-19 chest X-ray progression, benchmarked honestly across four metrics
- CIED Detection — Hybrid transformer and CNN pipeline for cardiac device detection, accepted as an ePoster at ACR 2026
- Chest X-Ray Multi-Model Diagnosis — Compared a medical vision-language model against general CNNs for 14-pathology classification
- ECG Anomaly Detection (VAE) — Unsupervised arrhythmia detection trained only on normal heartbeats
- RetailSense AI — Demand forecasting for a 54-store grocery chain, deployed twice over via Docker Compose and Kubernetes
- Voice-Driven B2B Lead Gen Agent — Real-time LiveKit voice agent that collects and confirms B2B leads
- Voice Chat Assistant · live demo — Full-duplex voice conversation app: speech-to-text, LLM reply, text-to-speech
- Copenhagen Cycling Ecosystem Analysis — Scraped and analyzed Copenhagen's bicycle retail market for underserved zip codes
- Job AI Assistant — 3-agent CV/interview coach pipeline with a self-healing validation loop
- Instagram Review Analysis — Fine-tuned Roman Urdu sentiment analysis on any public Instagram profile
- Voice-Cloned RAG Audiobook — RAG chatbot that answers in a voice cloned from a short audio sample
- SRS Generator — Multi-agent pipeline turning a project description into a UML-diagrammed SRS document
Khan MM, Raza A, Ahmad F, Mansoor A. "Generalizable detection of cardiac implantable devices using a hybrid transformer and convolutional neural network framework" ACR 2026 Annual Meeting, American College of Radiology, Washington DC · Accepted, abstract and poster
Raza A. "Inter-Lobe Boundary Channels and Localised Folding Repair for Expiration to Inspiration Lung CT Registration" Learn2Reg 2026 Challenge, MICCAI 2026, Strasbourg · Accepted, invited oral presentation and poster


