Fidel Tito

I'm a Bachelor in Mechatronics Engineering at Universidad Nacional de Ingenieria in Lima. Currently, I work as a Machine Learning Engineer at Qubika, where I specialize in deploying models into production, and implementing LLM-based applications .

Email  /  CV  /  Scholar  /  Linkedin

Research Projects

I'm interested in the areas of Software, Deep Learning, Computer Vision, Reinforcement Learning, ML Operations, and Image Processing. Additionally, I have experienced working in projects related to pose estimation, image segmentation, deep reinforcement learning, image alignment, control, and signal processing .

The research presents a Residual Vector Quantized Variational Autoencoder (RVQVAE) model for interpolating 2D keypoint motion in videos.

The research aimed to compare two advance control techniques to control the water level on a conical tank. Robust μ-synthesis and Gain Scheduling PID controllers were developed to facilitate this comparison, aiming to uphold level control.

The project sought to implement a mobile robot based on perception and control modules. Our perception module was implemented using UNet for image segmentation and DQN for pathfinding optimization, enabling a mobile robot to navigate indoors using a Jetson Nano.

JCA and ADA blocks were presented, and were responsible for reasoning about the coordination among the joints and appearance dynamics throughout the athletes’ performances. They captured fine-grained temporal depdendencies and long-term coarse-grained relations. Also, a new G-ExPose (Generalized ExPose) dataset was introduced.

A virtual refereeing network was proposed to evaluate the execution of a diving performance. This assessment was based on visual clues as well as the body joints sequence of the action video. Also, a new ExPose: annotated dataset of Extreme Poses was introduced.

A stratospheric balloon carrying sensors for weather monitoring was implemented. This implementention require sensor programming, signal processing, and the development of a real-time balloon movement monitoring interface.

Developer and ML Engineering Projects

Document Chat App is an application that allows you to ask questions related to AWS Documentation (or whatever markdown files you upload). It works with RAG and OpenAI models.

Web application built with FastAPI and PostgreSQL. This repository serves as a starting point for developing high-performance APIs with seamless database integration using Docker.

A ML classification model was trained. The model was deployed using Vertex AI Pipelines in GCP.

A virtual classroom was developed. It included frontend and backend using techonologies such as React, Redux, MongoDB, Express and NodeJS.

Miscelanea

Video instruction to deploy ML models using VertexAI GCP.

This website's style has been borrowed from Jon Barron's site' .