Open to AI/ML roles & research collaborations
Based in Linz, Austria
Ahmed
Mohammed
Computer Vision · Diffusion Models · OOD Detection
AI/ML engineer and entrepreneur. I build systems that work in the real world — from research to production.
4+
Years in AI/ML
99.03%
OOD Detection AUROC
+18.8pp
Improvement over baseline
<1 min
Report generation
~/faultrix $ python analyze.py --input site_photos/
Loading 42 images · GPU: RTX 4090
✓ Defect detection · AUROC: 99.03% ← SOTA
✓ Generating ÖNORM report... AES-256 encrypted
→ QC report ready <1 min
~/faultrix $
PyTorchDiffusion ModelsComputer VisionOOD DetectionCUDAPython
// 01 — About
AI/ML Engineer —
research that ships.
I'm an AI/ML engineer who completed my M.Sc. in Artificial Intelligence at JKU Linz, working under Prof. Sepp Hochreiter — the inventor of LSTM.
My work spans the full stack: from novel training methods for diffusion models and out-of-distribution detection in academic settings, to shipping production-grade CV pipelines that catch defects at 98.4% accuracy.
I founded Faultrix to turn this research into a real product — AI-powered construction quality control that generates ÖNORM-compliant reports in under a minute. Open to research collaborations and senior AI/ML roles.
Location
Linz, Austria
Supervisor
Prof. Sepp Hochreiter
Company
Faultrix.com — Founder
Open to
AI/ML Roles · Research Collaborations
Education
M.Sc. in Artificial Intelligence
Oct 2024 – Mar 2026
Johannes Kepler University Linz
Thesis: Conditional Diffusion Models as Generative Classifiers for Out-of-Distribution Detection — 99.03% AUROC on CIFAR-10
B.Sc. in Mechatronics Engineering
Feb 2015 – Jan 2018
Eastern Mediterranean University, Cyprus
Thesis: SCARA robotic system for dynamic object tracking (real-time control + sensor integration)
Skills & Stack
Deep Learning
PyTorchPyTorch LightningDiffusion Models (DDPM/UNet)CNNsTransformersRNNsLLMs
Computer Vision
OOD DetectionObject Detection (YOLOv8)Defect DetectionImage Classification
Research & Experimentation
HydraExperiment DesignAblation StudiesAUROC/FPR95Monte CarloCross-Validation
ML Engineering
End-to-End PipelinesLLM InferencePrompt EngineeringStructured Output ExtractionOpenAI API
Infrastructure
DockerGitLinuxREST APIsTypeScriptNext.jsConvexCUDAPythonOpenCV
AI Tools
GitHub CopilotChatGPTWeights & Biases
Languages
Arabic
Native
English
B2 · Professional
German
B1 CEFR · improving
// 02 — Projects
Research that ships.
⬡ Live SaaS · Founder
Faultrix
AI-powered construction quality control SaaS. Analyzes building photos and generates ÖNORM-compliant reports in under 1 minute — SHA-256 evidence chain, DSGVO compliant, AES-256 encryption. Built solo from zero to production.
faultrix.com ↗<1
min per report
ÖNORM · DSGVO compliant
Master's Thesis · JKU Linz
OOD Detection Framework
Conditional diffusion models as generative classifiers achieving 99.03% AUROC for robust out-of-distribution detection. Introduced class-conditional separation loss improving performance by 18.8 percentage points. Applied to industrial quality control with multi-head conditioning for structured manufacturing data.
99.0% AUROC (CIFAR-10)
DiffusionOOD ↗+ InkjetOOD · HF Weights
PROFACTOR GmbH · Zer0P
Industrial Anomaly Detection
YOLO + diffusion model pipeline for inkjet print quality control on building components. Open-source: YOLOv8 detector (0.950 mAP@50) + CDM scorer, pretrained weights on HuggingFace.
98.4% accuracy
InkjetOOD ↗+ HF Weights · Report PDF
Karunya University · BCI Research
EEG Signal Classification
LSTM · Bi-LSTM · GRU architectures for motor imagery classification in brain-computer interface applications. Hyperparameter optimization across multiple architectures.
// 03 — Experience
Where I've worked.
Jul 2025
Present
Founder & Full-Stack Developer
Faultrix · Linz, Austria
Built AI construction analysis platform solo from zero to production. Python, Next.js, Convex, OpenAI API, Docker, Clerk, Cloudflare R2, Stripe. ÖNORM-compliant reports with SHA-256 evidence chain and AES-256 encryption.
Active · faultrix.com
Dec 2024
Mar 2026
ML Researcher
JKU Machine Learning Institute · Linz
Thesis: "Conditional Diffusion Models as Generative Classifiers for Out-of-Distribution Detection" under Prof. Sepp Hochreiter.
Apr 2024
Nov 2024
Machine Vision Researcher
PROFACTOR GmbH · Steyr, Austria
YOLO + diffusion model pipeline for industrial defect detection. 98.4% accuracy in real-time production environment. Part of the Zer0P project — funded by the Government of Upper Austria — focused on zero-defect inkjet printing on building components.
98.4% accuracy
Aug 2023
Oct 2023
AI Research Intern
Karunya University · India (Remote)
RNN/CNN architectures for EEG motor imagery classification. Hyperparameter optimization across LSTM, Bi-LSTM, and GRU models.
Jan 2021
Present
AI & Programming Tutor
Freelance
Tailored training in Python, ML, and Deep Learning. Mentoring on portfolio projects and GitHub best practices.
// 04 — Research
Reports & Thesis
Master's Thesis · JKU Linz · 2026
Conditional Diffusion Models as Generative Classifiers for Out-of-Distribution Detection
Novel approach using conditional diffusion models as generative classifiers for robust out-of-distribution detection. Introduced class-conditional separation loss improving AUROC by 18.8 percentage points. Applied to industrial quality control with multi-head conditioning.
99.03%
AUROC · CIFAR-10
+18.8pp over baseline
Separation loss weight λ — key contribution
ROC curves — CIFAR-10 vs 5 OOD datasets
ID vs OOD score distributions per class
Novel contribution · Class-conditional separation loss
L_total = L_diffusion + λ · L_sep
L_sep = −MSE(pred_c=0, pred_c=1) ← widens ID/OOD reconstruction gap
λ = 0.02 → 99.03% AUROC (↑ from 80.3% baseline, +18.8pp)
vs published one-class baselines · CIFAR-10 airplane (one-vs-rest)
| Method | Type | AUROC |
|---|---|---|
| OC-SVM | One-class | 63.0% |
| Deep SVDD | One-class | 61.7% |
| DROCC | One-class | 81.7% |
| CSI | Contrastive | 89.8% |
| PANDA | Pretrained + OC | 95.4% |
| Mean-Shifted CL | Contrastive | 97.5% |
| Binary CDM (ours, λ=0.02) | Generative | 99.03% new |
Industrial Technical Report · PROFACTOR GmbH · 2024
Diffusion-Based Multi-class Defect Detection: A Generative Approach to Industrial QC
Production-oriented industrial quality control pipeline combining YOLO-based feature extraction with diffusion modeling for robust multi-class defect detection on inkjet-printed building components. Includes cross-domain and per-feature evaluation protocols used in real manufacturing settings.
98.4%
Defect Detection
Real-time production setting
ROC curves — inkjet CDM vs baseline, 5-fold CV
// 05 — Writing
Thinking out loud.
Deep-dives on research decisions, production lessons, and the gap between papers and products.
How I Hit 99.03% AUROC on OOD Detection Using Conditional Diffusion Models
My Master's thesis at JKU Linz (supervised by Prof. Sepp Hochreiter) introduced class-conditional separation loss into conditional diffusion…
Shipping Faultrix: What I Learned Building an AI SaaS From Zero in 5 Months
I went from a research background (diffusion models, computer vision) to shipping a production AI SaaS — Faultrix, an AI-powered constructio…
Diffusion Models for Industrial Defect Detection: 98.4% Accuracy at PROFACTOR GmbH
How I combined YOLOv8 feature extraction with conditional diffusion models to achieve 98.4% defect detection accuracy on inkjet-printed buil…
// 06 — Contact
Let's build something extraordinary
Open to AI/ML roles, research collaborations, and product partnerships.
ahmed.mo.0595@gmail.com