Deborah Danieli

Master’s Graduate in AI and Machine Learning. Experienced in developing AI-driven solutions for vehicle testing and structural health monitoring at SIEMENS. Passionate about transforming research into real-world impact through curiosity, collaboration, and continuous learning. Skilled in AI, software development, and data analysis, committed to creating efficient and trustworthy solutions.

Integration of Large Language Models (LLM) for Depression Monitoring in Social Media: 

• Designed an NLP system for early detection of depression by analyzing user post histories, used ChatGPT-4 APIs to generate psychological reports with enriched prompts and contextual metadata.
• 10% improvement in detection accuracy and interpretable results with Explainable AI.

AI-Driven Anomaly Detection and Severity Estimation in Vehicle EOL Testing: 

• Developed a Multitask Learning system combining segmentation (U-Net) and regression to identify anomalies and estimate severity.
• Improved segmentation accuracy (mIoU +15%) and reduced manual inspections.

Video Motion Magnification for Structural Health Monitoring: 

• Designed a real-time pipeline integrating classical Video Motion Magnification (VMM) with Neural architectures to detect micro-vibrations and structural anomalies.
• Enhanced clarity of visual outputs and reduced artifacts for more accurate structural health assessments.

Information Retrieval & NLP for Political Insight Extraction from Twitter: 

• Extracted political insights and user influence from 297K tweets during the 2020 U.S. elections using advanced NLP and IR techniques.
• Achieved alignment with official results and state-level trends through stance detection (T5 + LLaMA3) and semantic clustering (Sentence-BERT).
• Identified key political topics and top influencers within factions by building a user similarity network.

Optimization of ML Models for Dialogue Summarization: 

• Fine-tuned T5 and Pegasus models on the Samsum dataset to improve summarization of complex dialogues.
• Achieved significant improvements in summary quality, validated by ROUGE metrics.

Real-Time Data Analysis System with Apache Storm: 

• Developed a Python application for real-time analysis of physiological data, simulating a body sensor network.
• Implemented real-time ECG visualization, heart rate calculation, atrial fibrillation detection, stress level (HRV) analysis, fall detection, and movement tracking.
• Enabled continuous monitoring with low latency using Apache Storm and Kafka.

Noisy and Imbalanced Image Segmentation with Mask R-CNN: 

• Fine-tuned Mask R-CNN implemented in PyTorch and integrated with OpenCV for segmenting thin and complex objects in noisy, imbalanced datasets.
• Improved segmentation metrics by 12% through data augmentation and class imbalance handling techniques.

Violinist with 11+ years of conservatory and private training, winner of three first-place solo competition prizes, and experienced concert performer. Passionate about training and nutrition. 

I focus on achieving excellent results and believe in the power of teamwork and meaningful connections.