Filipa Soares

Assistant: design and rollout of a large‑language‑model copilot that lets stress‑analysis teams query finite‑element models and design specs Designed and developed scalable systems to automate financial document information extraction, integrating diverse models (GPT-4 Open, custom large language models, and Azure AI). Engineered robust Python-based logic for standardized JSON generation, dynamically handling missing fields, and reducing processing time by 70%. Developed end-to-end pipelines for deploying AI solutions on Azure, leveraging Databricks and Power Apps to ensure seamless integration, maintainability, and scalability of the system. Created modular Python codebases, implementing clean architecture principles, automated testing, and performance optimization to streamline information extraction workflows. Applied and deployed computer vision models for automotive clients (e.g., BMW) using PyTorch, achieving sub-1%error rates and a 20%improvement in image processing efficiency.

Developed and implemented state-of-the-art deep learning models (Mask R-CNN, Faster R CNN, and YOLO) for polyp detection and segmentation in medical imaging, processing datasets of 3000 images and improving inference speed by 10% on CUDA. This resulted in a significant 15%improvement in both accuracy and systemefficiency. Designed and implemented API integrations in C# to connect AI models with existing systems, ensuring smooth interoperability and extensibility for healthcare diagnostics platforms. Built reusable software modules for image preprocessing and data augmentation, enabling faster development cycles and consistent performance across diverse medical imaging datasets. Created and presented compelling proof-of-concepts to 50% of the company's clients, showcasing the potential of AI in healthcare diagnostics.

Achieved superior real-time performance with an Ensemble Learning classifier, using Principal Component Analysis (PCA) for feature selection and obtaining 96.53% accuracy in recognizing 20 different ADL and fall activities. Co-authored an article on "Inertial Data-Based AI Approaches for ADL and Fall Recognition," showcasing research contributions.