Projects

Delivered training sessions on generative artificial intelligence to oncologists from various hospitals.
Focused on how to implement generative AI to reduce administrative workload:
- Drafting emails (patient, insurance, doctor).
- Updating clinical guidelines.
- Writing medical reports.
- Analyzing and generating statistics.
Optimization of medical processes: clinical history review to accelerate decision-making.
Emphasis on the human-in-the-loop approach: AI should never make diagnostic decisions without human oversight.
Discussion on differences between assistant models and diagnostic models, as well as future advancements.
Three-hour workshop including practical examples (GPT, Copilot, etc).

Banking project aimed at reducing customer loss caused by the entry barrier into the investment world.
Development of an intelligent multi-agent system integrated into a chatbot to support decision-making.
Design of specialized agents:
- Manager Agent: distributed tasks among agents.
- RAG Agent: retrieved and analyzed documents from the bank.
- Web Agent: performed real-time internet searches.
System capable of answering both general investment questions (e.g., interest rates, risk concepts) and advising on the bank’s specific investment products.
Enabled extraction of the most suitable investment products for each client.
Fully developed on AWS, leveraging its cloud services for scalability and security.

Project focused on extracting key fields from PDF documents containing company incorporation deeds.
Required legal understanding to properly annotate ground truth and validate model performance.
Technical process:
- OCR to convert PDF into text.
- OpenAI API (GPT models) with advanced prompt engineering (zero-shot, few-shot, chain of thought).
- Queries split into smaller tasks to improve accuracy.
Main achievement: increased performance to 97% accuracy by subdividing tasks into multiple model calls.
Implemented in Python with field-by-field exact-match validation and automated iteration checks.
Final deliverables in JSON format using Pydantic for structured outputs.
Later, handled incremental client requests in production, analyzing new use cases and estimating requirements.

Conducted a state-of-the-art review (May 2024) of large language models deployable on local devices.
Identified key advantages:
- Enhanced privacy and data control.
- Independence from internet connectivity.
- Usability in low-connectivity environments.
- Reduced cloud computing costs.
- Hardware-specific customization and optimization.
Studied optimization techniques to make LLMs lighter and more efficient, including:
- Pruning – removing redundant weights.
- Quantization – reducing numerical precision for faster inference.
- Knowledge distillation – transferring knowledge from large to smaller models.
- Low-rank adaptation (LoRA) and fine-tuning strategies.
Findings were used to design tailored proposals for clients requiring local deployments.

Mentored junior profiles since the beginning of my role at the company.
Explained technical concepts to facilitate onboarding and ensure proper task execution.
Provided continuous support, resolving doubts and guiding their work.
Combined mentorship responsibilities with active participation in high-level project development.

Data project focused on preparing, transforming, and structuring large datasets.
Data cleaning and transformation performed with PySpark.
Delivered insights by integrating data into MicroStrategy dashboards, allowing the client to analyze capital flows and monetary fluctuations across different stages of their internal tool.
Required strong understanding of:
- Banking regulations.
- The bank’s internal systems.
- Client-specific requirements and workflows.

Software development project aimed at automating processes in telecommunications networks to reduce human error.
Implementation of automation solutions using Python, with focus on communication networks and router operations.
Additional responsibilities included team training:
- Conducted onboarding sessions for new team members.
- Explained technical concepts to facilitate integration into the project.
- Provided continuous support to resolve questions and ensure team efficiency.