Ramon Perez

Built performance benchmarking and visualization tools for local LLM deployment in robotics contexts. Created robobench (Python/C++ library for measuring real-world LLM performance) and roboviz (real-time visualization of token flow through CPU architecture using Python/Vue.js). Presented research on C++ LLM integration at CppNow 2025. Rebuilt Jan.ai's documentation architecture, created MCP integration tutorials, and conducted user research to shape product strategy. Currently working on simulation platform and data engine infrastructure for the Asimov robotics team.

Actively contributed to Seldon's suite of dev tools with emphasis on the paid machine learning inference products, crafted educational content for the OSS tools, delivered talks at conferences on different ML topics, and engage with our developer community while coordinating the MLOps World Meetup group events, fostering stronger connections and support within our network.

Created Decoded's initial data engineering platform on AWS, and developed four machine learning (ML) applications for language translation, churn prediction, fraud detection, and content recommendation. I led the creation of our MLOps engineering program (see it here) and contributed to our React-based presentation tool. I still teach for Decoded's biggest clients.

During my tenure at INSEAD, I worked with a team of outstanding researchers to develop algorithms for analyzing company structures and culture. Our main project focused on generating hierarchical organizational structures (vertical or horizontal) by analyzing historical job advertisements from target companies and similar organizations. We leveraged various natural language processing techniques alongside supervised, semi-supervised, and unsupervised machine learning methods. Additionally, I developed a Python package implementing Latent Dirichlet Allocation to automatically extract and select meaningful topics from company reviews on platforms like Glassdoor. This package not only identified key themes but also provided quantitative cultural measures for organizations based on the discovered topics, offering valuable insights into organizational culture and structure.

At GFLEC, I led multiple research initiatives focused on retirement and financial education. I developed and deployed a comprehensive MTurk survey examining millennials' financial education seeking behaviors, while simultaneously conducting extensive literature reviews and expert interviews with 20 policymakers and retirement specialists to understand phased retirement programs. Using the Health and Retirement Study dataset, I built statistical models in Stata and R to analyze patterns between future workforce participation and baby boomers' gradual retirement trends, providing valuable insights into retirement planning and financial literacy.

Created ArcGIS maps to visualize consumer behavior data, while concurrently conducting sentiment analysis using SPSS and Python, and orchestrating A/B tests to assess the efficacy of various data collection strategies like email, mail, and point-of-sale.