About me

vituri.vituri@gmail.com / https://github.com/vituri

EDUCATION

Doctor of Philosophy – Topological Data Analysis
Universidade Estadual Paulista (Unesp) / Ohio State University (OSU) 2016 — 2020

  • I spent six months at the Ohio State University under Facundo Mémoli supervision.
  • We generalized the Vietoris-Rips and Cech constructions using “motifs”: families of graphs that can be attached to metric spaces (more generally, networks) and generate simplicial complexes. As a corollary, we obtain several clustering methods on graphs.
  • The resulting work is “Motivic constructions on graphs and networks with stability results” and can be found here.

Master Degree – Algebraic Topology
Universidade Estadual Paulista (Unesp) 2014 — 2015

  • I studied a certain subgroup of the fundamental group, called “Gottlieb group”, and proved some results about its properties. The dissertation can be found here.

Bachelor in Mathematics
Universidade Estadual Paulista (Unesp) 2010 — 2013

WORK EXPERIENCE

Head of Intelligence

Argus Solutions Jan 2020 – Jan 2025

I founded the area of Data at Argus, and my role was crucial as the company went from a small-sized tech to a multinational. My team now has more than 12 people as diverse as developers, data scientists, engineers and mathematicians. I act as a manager, code reviewer, project manager, machine learning engineer, DevOps, R/Julia developer and database administrator. I can do a bit of everything.

  • We went from hundreds of Excel files to a MariaDB database on AWS. I designed the database, its indexes and applied some normal forms.

  • I wrote dozens of R/Julia scripts to read and write data from several sources and created a pipeline of information that ran smoothly

  • I automated the generation and mailing of thousands of daily/weekly reports using RMarkdown and Sendgrid, which previously was done by 3 people.

  • I created several analytics dashboards using the shiny framework, in place of a former Tableau dashboard. I also developed the main platform used by the team to analyze photos/videos.

  • We created a machine learning model that predicts if a driver is going to sleep in the next hour, using an ensemble of models with tidymodels.

  • I created a webserver in Julia using Oxygen.jl to read data from several APIs and write they in our databases. A previous version of this code was written in R and each iteration took almost 4 minutes; it was reduced to 5 seconds in Julia using some paralellism magic.

  • I participated in the creation of many computer vision algorithms to detect cellphones, drowsiness, potholes and gestures using keras and YOLO.

  • I interviewed and hired people, managed the team and developed the abilities we needed to finish our projects.

TOOLS AND SKILLS

R

  • tidyverse: I use several of these packages everyday; I am “fluent” in dplyr, purrr, tidyr, stringr, tibble and many others. I prefer to use echarts4r instead of ggplot2. With dbplyr, I can translate many complicated transformations to SQL in a painless way.

  • shiny: I explored a lot of the shiny ecosystem since 2018: from shinydashboard to the new bslib, using shinyWidgets and toastui. I can create dashboards, CRUDS, user interfaces, display photos/videos and maps. Async and parallelism are a bit painful but can be done with future and promises.

  • I am also very familiar with igraph to manipulate graphs, leaflet to plot maps, httr2 to make gets and posts, plumber to create APIs, reticulate to interface Python.

  • I can create packages, documentation and tests. Unfortunately, my best packages are private but a public example can be found here.

Julia

Coding in Julia is a joy and make R/Python look like unpolished old languages used by barbarians.

  • I started learning Julia to write some Topological Data Analysis algorithms and the result was impressive, even with my initial poor skills.

  • I am the owner of the JuliaTDA organization which aims to bring the Topological Data Analysis ecosystem to Julia. I wrote the Mapper and Ball Mapper algorithms in Julia and the documentation in Quarto, and later also wrote the ToMATo algorithm.

  • Building JuliaTDA, I was able to get familiar with parallelism, code piracy, structs, multiple dispatch, profiling (memory allocations and algorithm efficiency) and plotting tools like Plots and Makie.

  • The book “Hands-On Design Patterns and Best Practices with Julia” is my bible, and I am rewriting all my previous code using what I learned there.

  • Since I like the tidyverse so much, I am now part of the TidierOrg: a 100% Julia organization inspired by R data tools. I am the main contributor to TidierIteration.jl, a Julia version of R’s purrr.

  • Since I like Quarto so much, I created a package that helps writing Julia documentation with it: QuartoDocBuilder. Its own documentation is written in Quarto!

AWS

  • I can create and manage EC2 instances, use S3 to storage and query data, create databases in RDS and create dashboards with metrics.

  • I created a facial recognition system of unlabeled drivers using S3 and Rekognition.

  • Recently I moved several R/Julia/Python scripts from EC2 to ECS services, providing a reproducible and isolated environment for each code. While doing this, I learned how to create images with Docker, building them with EC2 Image Builder and storing it all in ECR.

Python

  • I can read code and port it to another language when necessary; I never had the need to write pure-Python code, but can learn it easily if needed. Using R’s reticulate and Julia’s PythonCall.jl, I was able to use many Python packages elsewhere.

Publishing

  • I used a lot of RMarkdown to generate a wide range of dashboards, reports and analysis. Now I prefer to use Quarto because of its many cool features (sites, blogs and dashboards) and easiness to use with Julia.

  • I am writing a book called Topological Data Analysis with Julia while I implement several TDA algorithms in JuliaTDA.

Talks, workshops and papers