Anar Amirli

Working on evaluating and mitigating social biases in multimodal large language models (LLMs).

• Developed an ante-hoc, concept-based explainable AI (XAI) model with Graph Neural Networks for medical imaging (skin cancer diagnosis); improved accuracy by ~3% vs. baseline CBMs. • Conducted extensive research on foundation models (e.g., CLIP, MedCLIP) to benchmark and analyse concept-based explainability in medical imaging. Team: Interactive Machine Learning

• Developed and deployed a self-supervised anomaly detection system (FastAPI, Docker, AWS) for Schott AG manufacturing lines, boosting anomaly localisation accuracy by 13% with post-hoc XAI methods and reducing defect-related downtime. • Fine-tuned transformer-based LLMs (e.g., T5, BART) to generate incident reports from telemetry sensor data to assist early incident assessment with automated reports. Team: Smart Service Engineering

• Developed a multimodal-to-image translation pipeline using U-Net-style generators and GANs, achieving 91–99% reconstruction accuracy and enabling near real-time topology optimisation of 2D/3D structures. • Deployed real-time ML models (Docker, Flask/FastAPI) to replace heavy simulations, improving efficiency and scalability. Team: TESLAB

• Contributed to the development of a real-time speech recognition system for the Universal Virtual Simulator Project at the Azerbaijan National Aviation Academy. • Engineered audio features (spectrograms, MFCCs) and trained LSTM/HMM models on cockpit command data.

• Designed and implemented a deep learning model for ball position estimation in football, helping tracking cameras handle occlusions. • Processed and analysed large-scale football streaming data of league games spanning multiple seasons, including extensive data cleaning, feature engineering, and visualisation. Team: ImageLab

• Developed multi-feature Android apps with backend integration (HTTP APIs to Java interfaces) and contributed to the mobile app development for Opal Transfer LTD.

• Thesis Grade: 1.0 (German scale) • Recipient of the Deutscher Akademischer Austauschdienst (DAAD) Full Graduate Scholarship Comprehensive theoretical and applied coursework in Data Science, AI, Optimisation, and Machine Learning, taught by leading research institutes in Germany, including DFKI, Max Planck Institute, and CISPA.

• Grade: 1.3 (German scale) • Graduation with Honours • Government Scholarship for Academic Excellence Activities: • ICT and Robotics team member