Cyrine Chaabani

Building large industrial foundation models for the automotive industry, applied to CAD/CAE workflows.

Helping operators of industrial real-estate across South Africa forecast, simulate and orchestrate distributed energy assets using ML forecasting models and Agentic AI.

Deployed LLM-based agentic compliance workflows (70% faster audits) and built backend APIs embedding LLM reasoning for cloud-deployment optimization.

Shipped an LLM + RAG chat assistant (95% resolution rate) and an adaptive customer-journey personalization engine that boosted conversion on financial products by +60%. Built fine-tuning and data-scraping pipelines, set up live agentic-architecture monitoring with Langfuse, and led AI hiring while mentoring junior engineers.

Published on explainable generative AI for energy, led work on multi-agent optimization with transformer- and RL-based architectures, and taught product management, ideation and rapid prototyping to 50+ graduate students.

Developed grapheme-to-phoneme models for German and its dialects (+10% accuracy over the previous architecture). Built a real-time sign-language detection system (87% accuracy) using graph learning with bias-aware attention and a specialized loss, and a multimodal deep-learning system for sign-language detection that won the Best Model prize at WMT22.

Leveraged ontologies to improve planning edge-cases for autonomous navigation of STR robots. Generated 4,000+ synthetic audio/visual training examples via Nvidia Isaac Sim and built data ingestion and evaluation tooling for continuous benchmarking.

Delivered AI solutions across tabular classification, document understanding, automated information processing and visual defect detection (XGBoost, CNNs, LSTMs, ensembles, NLP). Acted as lead ML engineer on 5 multi-disciplinary R&D teams for clients including Henkel, OSRAM and the Ministry of Energy in Dubai, with a 100% delivery success rate.

Built real-time RGB-D object detection (96% accuracy) and developed the vision component of a prototype robot that detects, grabs and relocates target objects (CNN-based detection, pixel-to-world coordinate translation, SLAM-based navigation).