Nayeemuddin Mohammed, AI Infrastructure / MLOps Engineer

Two years of production DevOps and cloud work at Tata Consultancy Services (AWS, Terraform, Kubernetes, GitLab CI/CD). At TH Deggendorf I applied that foundation to AI systems: real-time computer vision for factory-floor defect detection, and generative pipelines that turn product text into catalogue-ready imagery.

Experience

• Research Assistant, AI & Computer Vision, Technische Hochschule Deggendorf, Feb 2026 to Jul 2026. Delivered a YOLOv12 assembly-inspection pipeline on Raspberry Pi 5 spanning 16 product variants, reaching 95% mAP@50, precision, and recall, and replacing manual visual inspection with real-time multi-camera defect detection. Cut model inference under 500ms, with the full edge-to-MES round trip (detection through the REST API to a recorded result) landing under one second per unit. Designed the REST API (Python/FastAPI) and offline-queuing layer linking edge inference to Tulip MES, producing zero-duplicate, audit-ready quality records with full edge-to-production traceability.
• Research Assistant, AI & Statistical Modelling, Technische Hochschule Deggendorf, Sep 2025 to Dec 2025. Designed and ran CCD-RSM, D-optimal, and mixed factorial DOE experiments across additive manufacturing and robotics domains. Developed R/Python pipelines combining ANOVA and ML-based optimization to replace manual statistical analysis. That pipeline is now used to teach DOE methodology to THD students.
• DevOps Engineer, IoT & Cloud Infrastructure, Tata Consultancy Services, Mar 2023 to Feb 2025. Replaced manual deployment steps with GitLab, Jenkins, and SonarQube pipelines for IoT firmware and backend releases, cutting deployment time from 48 hours to under 30 minutes and moving release cadence from bi-weekly to on-demand. Provisioned and managed Terraform-based infrastructure on Kubernetes (Amazon EKS), containerizing and running 50+ microservices across 300+ pods with Docker in 4 lifecycle environments, monitored through custom PostgreSQL-backed KPI dashboards in Grafana. Deployed AWS IoT Core, Lambda, and IoT Analytics for an event-driven data architecture over MQTT and HTTP, sustaining 99.99% uptime across 10,000+ active edge devices and 2M+ daily events.

Projects

• AI-Based Automated Assembly Inspection with MES Integration. Three cameras feed a YOLOv12s ONNX model on a Raspberry Pi 5. Results post straight into a Tulip MES, with an offline queue so records stay complete and duplicate-free.
• Automated Defect Detection with Edge AI. A two-stage YOLOv11n pipeline for colour and missing-piece defects, quantized to INT8 and running on-device on an ESP32-P4.
• Multimodal Generative AI Pipeline for Fashion. Four models in sequence: Helsinki-NLP translation, Phi-3 Mini for prompt writing, FLUX.1-schnell with a LoRA for generation, and CLIP scoring to pick the keepers.
• Fetal Health Classification, ML Pipeline. Logistic Regression, Random Forest, and SVM compared, with SMOTE applied inside the training pipeline so nothing leaks into the test set.
• Cost-Aware Demand Forecasting. ARIMA, SARIMA, and Prophet backtested on DataCo order data with an expanding window. Plain ARIMA won both categories on total inventory cost, because SARIMA's seasonal term overfit the smaller backtest windows.
• Zollhof Impact Digital Twin. Built with my team in TypeScript at the Claude Impact Lab Hackathon in Nuremberg. One Claude Opus call returns a Zod-validated, multi-dimensional impact profile against the IMP framework, IRIS+ themes, and UN SDG targets, shipped live as a Vercel serverless API.
• Queueing Simulation and Optimization System. Discrete-event M/M/c call-center simulation in SimPy that finds the lowest-cost staffing level still meeting the wait-time SLA, cross-checked against Erlang C.
• ARIMA Time-Series Forecasting. Single-file R pipeline for ARIMA forecasting of COVID-19 case data: order selection, stationarity tests, residual diagnostics, and a 30-day forecast with confidence bands.
• Catapult DOE. Catapult DOE study in R: measurement system analysis plus a full factorial run sheet, analysed in a tidy Quarto model.
• Paper Helicopter DOE. Paper-helicopter DOE in R: full and fractional factorial designs on flight time, with ANOVA, model diagnostics, and optimal factor settings.
• The Future of Manufacturing in 2040. Quarto/LaTeX case-study report on what manufacturing looks like in 2040, with APA-numeric citations and a rendered PDF.

Skills

• MLOps & Infrastructure: Kubernetes, Terraform, Docker, AWS, GitLab CI/CD, Jenkins, MLflow, Grafana, IoT Core, Lambda, CloudFormation
• ML & Computer Vision: PyTorch, YOLOv12, OpenCV, Hugging Face Transformers, Scikit-learn, TensorFlow, Keras, CLIP, LoRA fine-tuning, Diffusion models
• Programming: Python, FastAPI, R, NumPy, Pandas, PostgreSQL, TypeScript, JavaScript, HTML / CSS
• Edge & Embedded: Raspberry Pi 5, ESP32, MQTT, CoAP, REST API

Education

• M.Eng in Applied AI for Digital Production Management, Technische Hochschule Deggendorf, Mar 2025 to Present. Grade: 1.8
• B.Tech in Mechanical Engineering, SRM Institute of Science and Technology, Jun 2019 to May 2023. CGPA: 9.03
• English Native, German A2

Contact

nayeemuddin.mohammed@stud.th-deg.de, +49 15563598740