Joan Capell

- Built CallMazeSolver, an AI voice assistant that places phone calls on a user's behalf to handle slow administrative workflows such as insurance interactions and other customer-service processes

- Built ChessAlterEgo, a deep learning project that models and reproduces human chess playing style using deep learning

- Developed an OCR and computer vision based pipeline to automatically validate PCs from images and extracted device information.

- Built an end-to-end email automation workflow using Transformers, XGBoost, and GPT models for classification, response generation, and ticket creation

- Developed voice agents with Gemini models and built the data backbone for semantic retrieval and RAG workflows

- Led containerization, CI/CD, testing, and deployment practices for AI applications in production environments

- Mentored team members on the effective use of AI coding assistants to accelerate development and improve code quality

- Trained and benchmarked Transformer-based models on AWS for enzyme and chemical reaction classification, and packaged them into tools usable by research and chemistry teams

- Developed generative AI models to design novel enzymatic sequences for reaction-specific discovery workflows

- Built a RAG system over scientific literature using Llama and MongoDB, enabling natural-language access to internal research knowledge

- Containerized applications with Docker and implemented CI/CD pipelines with automated Pytest coverage