AI Engineer crafting intelligent systems at the intersection of Mathematics, Machine Learning, and Deep Learning. Based in Munich, Germany.
I'm an AI Engineer and Applied Mathematician based in Munich, Germany, currently completing my Master's in Mathematics in Science and Engineering at the Technical University of Munich.
My expertise spans the full ML lifecycle — from translating business problems into technical specifications, to building production-grade pipelines, to deploying uncertainty-aware models at scale. I've delivered solutions for clients in healthcare (AI Radiologist), insurance, automotive (Audi AG), and transportation policy.
I thrive at the intersection of rigorous mathematics and practical engineering, believing that the best AI systems are built on strong theoretical foundations and clean, reproducible code.
Uncertainty Quantification for GNN Surrogates in Traffic Policy Modeling
Identification of Neural Networks (Prof. Massimo Fornasier)
Technical University of Munich
Aligarh Muslim University
Real diagrams from my thesis, research, and project work
Production ML workflow: from business requirements to deployed, monitored models.
Graph Neural Network replacing expensive transport simulations with uncertainty quantification.
Formalizing when neural network parameters can be uniquely recovered — symmetry breaking strategies.
Comparative model evaluation across key metrics for the AI Radiologist prototype.
End-to-end deep learning system for medical image analysis. Built dataset pipeline, experimented with CNN architectures, and delivered a demo-ready prototype for stakeholder review at BP-ITCS.
Graph neural network models that replace expensive agent-based simulations to predict policy effects in unseen cities. Features uncertainty quantification and multi-fidelity learning.
ML solution for an insurance client featuring iterative model selection, calibration, thresholding, and performance improvements aligned with business KPIs and risk metrics.
Fine-tuned BERT for multi-class sentiment analysis on product reviews. Implemented data augmentation, attention visualization, and achieved 94% F1-score with optimized inference pipeline.
Built a YOLOv8 detection pipeline for autonomous driving scenarios. Custom-trained on urban scenes with data augmentation, achieving real-time inference at 45 FPS on edge devices.
LSTM and Transformer-based time series model for predicting energy consumption patterns. Implemented multi-horizon forecasting with attention mechanism and temporal fusion transformers.
Collaborative filtering and neural collaborative filtering approach for product recommendations. Implemented hybrid model combining content-based and interaction-based signals.
Interactive Tableau dashboards with cross-filters and drill-downs. Visualized flow and volume fields in ParaView using streamlines, slices, and isosurfaces for scientific data.
Power BI dashboards for a car-parts manufacturer exposing critical metrics: time savings, ROI, defect rate, and perfect order rate. Earned stakeholder praise for intuitive UI/UX.
Migrated Audi's ad-hoc spreadsheet workflows to structured VBA modules. Built a macro-driven UI that reduced data-validation time by ~90% with unified ownership and approval views.
Orchestrated complete ML pipelines on real datasets covering supervised and unsupervised learning. Implemented SVM, classification, regression, and clustering with rigorous hyperparameter search.
Built an autoencoder-based anomaly detection system for IoT sensor data streams. Integrated statistical process control with deep learning for multi-modal anomaly scoring.
Implemented StyleGAN2-ADA for high-resolution image generation with adaptive discriminator augmentation. Achieved FID score of 12.3 on custom dataset with progressive growing and style mixing.
Built a framework for efficient model tuning using Gaussian Process-based Bayesian optimization. Reduced search time by 70% compared to grid search while finding better configurations.
Designed a RAG pipeline using LangChain and vector databases for domain-specific Q&A. Implemented chunking strategies, embedding fine-tuning, and re-ranking for improved answer quality.
Trained PPO and DQN agents for complex control tasks in simulation environments. Implemented reward shaping, curriculum learning, and multi-agent coordination strategies.
Built a cross-attention model fusing tabular, image, and text data for patient outcome prediction. Implemented late fusion and early fusion architectures with attention-weighted modality gating.
Implemented federated averaging with differential privacy for decentralized model training across institutions. Achieved 96% of centralized accuracy with zero data sharing.
DeepLearning.AI · Coursera
September 2024 View Certificate →DeepLearning.AI · Coursera
September 2024 View Certificate →Udemy
August 2024 View Certificate →DataCamp
April 2022 View Certificate →DataCamp
April 2022 View Certificate →I'm always open to discussing new opportunities, AI projects, or collaborations. Feel free to reach out!