I created the AI Mini Scientist project, an automated solution that performs intelligent search, extraction, and organization of scientific articles from the PubMed database. Built with Python and Pandas, it integrates PubMed’s API, prompt engineering, and orchestrates interactions using LangChain and LLMs for summarization and classification of results.
The system automatically generates structured reports and scientific summaries that feed a large-scale virtual library developed for a healthtech startup aiming to implement medical query tools in public health centers using only peer-reviewed scientific evidence. This solution enhances literature review processes, delivers actionable insights for innovation and clinical decision-making, and significantly reduces research time while improving the accuracy of retrieved information.

I designed and implemented a local multi-agent AI system for business automation, built with Ollama, OpenWebUI, and LangChain, fully integrated with SAP via RPA technologies (Python, Power Automate).
The first agent uses prompt engineering and natural language processing (NLP) to serve as a natural language interface, identifying user intent, extracting relevant query parameters, and converting complex requests into structured automation instructions.
The second agent receives these semantic instructions and translates them into executable scripts that perform real-time data extraction, filtering, and analysis directly from SAP, as well as managing business automation flows.
The entire system is designed with a strong emphasis on security, running fully offline, and applies machine learning and LLM techniques to improve request comprehension and system robustness across various corporate scenarios.

I developed a fully offline automatic translation tool for PowerPoint presentations, powered by Hugging Face’s mBART model. I fine-tuned the model using a proprietary parallel corpus and an internal corporate glossary to ensure accurate translation of text boxes and tables. The solution prioritizes security by running entirely locally without transmitting sensitive data, and it leverages advanced NLP techniques to adapt translations to the company’s specific business context.