Contributed significantly to designing, developing, and testing an Agentic Retrieval-Augmented Generation (RAG) system• The system crafts expert insights for complex queries in domains such as Chemicals, Consumer Products, Oil and Gas, Utilities, etc. • Created a vector database to process and index over 35,000 documents, including relevant metadata.• Evaluated various embedding algorithms, reranking methods, and similarity cutoffs to achieve optimal retrieval performance.• Designed and developed an Agent that decides which tools to call based on user queries, including the RAG system.• Developed a WebSocket API to integrate the Agent's capabilities with the front end.• Collaborated with the team to reduce system latency from 35s to 8s by sending the response through the WebSocket connection directly from the tool instead of the Agent processing the tool's response.• Developed an automated data pipeline that fetches data from an API and stores it in a vector database to update the RAG system with the latest information from the primary database.• Designed and implemented a feature to analyze chat history and rank topics by relevance to the conversation. This ranking system automatically recommends the most suitable analyst for the user to discuss the given topic, eliminating the need for manual matching previously required by the CX team.• Designed a system monitoring mechanism using LLM-as-a-judge techniques to ensure answer relevance, groundedness, and conciseness using Langfuse and DeepEval.• During the first phase of the project, it achieved 80.3% accuracy, 77% relevance, and an 87.3% average tone rating. This system was thoroughly evaluated by expert directors and PhD-level researchers using over 100 varied questions to calculate these average scores.• Responsible for deploying product features across development, staging, and production environments and troubleshooting issues.

• Collaborated with a software engineering team to develop a no-code application that accelerated data scientists' iteration speed by 4x during the development of advanced analytics projects for customers.• Developed a Python algorithm that combines machine learning models based on performance and cost. Applied it to a proof-of-concept with a digital bank, reducing costs by an estimated R$630k while maintaining model performance.• Automated a machine learning (ML) pipeline using Airflow, reducing model development time from one week to 2 hours.

• Spearheaded the development of a software solution that decreased the volume of queries made by a single client to a paid source by 3.81%, resulting in an estimated yearly savings of 1.7 million queries and R$3.4 million in total billing (R&D).• Maintained communication with 5 key stakeholders (CTO, Product Director, Project Manager, Sales, Customer Success teams), presenting and discussing project results.

• Innovated by creating a machine learning algorithm that reduced cache consumption by 86.7% and simplified the credit policy by 49.5%, using Python, C, Cython, and scikit-learn.• Engineered a Python-based application that combined structured (XML) and unstructured (text) data to generate nearly 10 comprehensive reports weekly, ensuring accuracy and facilitating informed decision-making in ML model development.