Ruanitto – LLM, RAG, LangChain

Enterprise leadership development platform combining professional coaching, AI-powered tools, and analytics to help organizations improve leadership, team performance, and employee development.

• Owned the evaluation pipeline for LLM-powered services, increasing guardrail coverage by 20%;
• Increased chatbot activation by 17% by using LLMs to generate pre-calculated conversation starters.

AI-powered multi-agent platform helping digital creators automate content creation, image and video editing, analytics, and other day-to-day creative workflows.

• Developed a multi-agent system using AI to automate and simplify creative workflows for digital creators;
• Built a modular and resilient agent architecture using LangChain, LangSmith, and proprietary LLMs;
• Spearheaded the LLM evaluation platform to detect regressions and prevent performance degradation.

AI-powered industrial maintenance platform helping factory teams streamline maintenance workflows, diagnose equipment issues, and improve the accuracy and efficiency of maintenance operations.

• Contributed to the first LLM-powered services for industrial maintenance, improving the speed and accuracy of maintenance workflows;
• Led RAG implementation using LangChain with GPT-4o and Claude to retrieve relevant maintenance information;
• Implemented LLM evaluation and monitoring using Weights & Biases and Promptfoo;
• Implemented document parsing with vision-language models to extract structured data from unstructured documents;
• Built an agentic web-scraping infrastructure using LLMs and Selenium to collect relevant images from the web.

Mobility technology platform providing vehicle access and rental services for app-based drivers, using IoT and data-driven solutions to improve driver safety and reduce accident risk.

• Played a key role in developing a driving-behavior model based on real-time IoT data from vehicle trackers, contributing to a 40% reduction in customer-caused accidents;
• Led development of the second version of the driving-behavior model, using PySpark for feature engineering and genetic algorithms for optimization;
• Built an internal library to simplify the data science team’s integration with commonly used field libraries;
• Improved CI/CD and established standardized practices for model training, testing, and deployment;
• Led development of a GitHub Actions-based interface for launching and monitoring model training;
• Developed an end-to-end Random Forest model to identify potential vehicle battery failures;
• Built a neural network to predict customer mileage over the following 30 days;
• Developed a parts demand forecasting solution combining linear models and neural networks;
• Implemented SHAP-based explainability for neural network models;
• Developed outlier detection models for business processes such as vehicle maintenance using AWS SageMaker, Athena, and Metabase;
• Built an automated, model-agnostic outlier detection pipeline supporting automated training and deployment with Python, Docker, and AWS SageMaker.