Projects
This page showcases some projects I have worked through my life.
Projects protected by NDAs are not shown on this page.
Most images and videos displayed here are merely illustrative and were not used on the actual projects.
For the CV
Radar Esportivo (Sports Radar)
A platform where soccer traders can easily access, on a single window, the best data available across the web. Radar Esportivo started as a side hustle and grew enough to need its own company (433Data) and a few months fully dedicated to it.
With a free tier without the need for accounts and a UI that puts speed and simplicity above all else, the product is a go-to spot for many Brazillian traders to use during games. Those who need even more can opt for the paid plan. It gives them the ability to follow multiple games simultaneously and custom metrics for premium championships.
Developed as a side hustle, the project is entirely serverless, scaling efficiently, and with low operation costs. It also needs no human intervention, being completely automated. Projected as a semi-passive income source after the initial coding, it only needs human intervention for new features and eventual maintenance.
- Cofounded
- 15,000 active users
- Half a million monthly page views
433Data
A branch from the Radar Esportivo project, this project named the company I currently own and aims to deliver state of the art stats for soccer games, teams, and players. Built using custom machine learning algorithms and validated against the most prestigious competitors, it will be a premium product with high acquisition cost.
As with any machine learning project acquiring adequate data and allowing for the model to automatically update itself has been a challenge, especially during COVID-19 times where games have been inconsistent in both schedule and players performances.
This project is currently on closed beta.
- Cofounded
- State of the art machine learning
- Automated Learning
Automated Stroke Detection on CT images
One person in six you will have a stroke during their life (if they live in Brazil). If the stroke diagnosed in less than 4 hours, the patient may fully regain their mobility, self-care, and social skills, progress sooner, or experience only a slight decline.
As 95% of the hospitals in Brazil do not have 24/7 radiologists (not even on-call), this system exists to help decision making on emergencies where there is an absence of those professionals. The system can analyze and detect anomalies in brain CTs, and also inform the presence of blood, corroborating the safety of using tPA.
This project is a partnership with Hermes Pardini, one of the largest exam labs in latin america. Currently, in phase one, the system can only deal with strokes and hemorrhages (for now) and is in active development.
- Project Leader
- Life saving system
- Extremelly reliable serverless architecture.
Automatic serial reader for steel slabs
Through strategically positioned cameras, this software automatically detects, segments, recognizes, and groups serial inscription numbers (SINs) for steel slab stacks with variable sizes using computer vision.
Developed for Gerdau, one of the world's largest suppliers of steel, this software contributes to their cyber-physical space initiatives and automates a previously manual (and potentially dangerous) task of controlling their stock throughout the storage facilities even when they are in motion.
This project was a runner up at the Mercosul Science and Technology Awards.
- Project Leader
- Fully integrated
- Harsh operational environment
K-in-a-Tree
The Three-in-a-tree algorithm solves the following problem in polynomial time: given three fixed vertices in a simple finite graph, check whether an induced tree containing these vertices exists. In my Ph.D., I deal with a generalization of this problem, referred hence-forth as k-in-a-tree.
The k-in-a-tree algorithm checks whether a graph contains an induced tree spanning k given vertices. When k is part of the input, the problem is known to be NP-complete. If we fix k≥4, its complexity is an open question, although there are efficient algorithms for restricted cases such as claw-free graphs, graphs with girth of at least k and chordal graphs. We present mixed-integer programming formulations for this problem, and we show that instances with up to 25,000 thousand vertices can be solved in reasonable computational times.
- Ph.D. Thesis
- Gaphs
- Work in progress
Progen
A computer system for the daily programming of electro energetic operations. It can simulate hydroelectric power plants, considering constraints, defluences uncertainties, programmed interventions, hydraulic turbine efficiency curves (unit commitment), among other characteristics inherent to short-term scheduling;
It is also able to maximize the use of the water resources and pair both short and long term energy guidelines; All of this in a gorgeous easy to use interface that also automates the compliance with regulatory agencies.
Operational in more than 30 power plants, with 50 more planned.
- Architect
- Technology Evangelist
- Internal Consulting
Slag Detection
Through strategically positioned cameras, this software is capable of detecting, qualitatively, the amount of slag coming out of a Guza Steel blast furnace.
The output was used as part of a bigger optimization software.
- Project Leader
- Prototype Development
- Camera Specification for harsh operational environment
Corporal Prediction for Cattle
A common problem on beef cattle creation is knowing the perfect slaughter time frame, nowadays animals are considered ready in batches. Although most of them may be in the correct time frame, some may still be under or even over developed, havin less commercial value than the others.
This project used earring tags in conjuction with both normal and 3D cameras near the feeding spots to collect data on the animals. This data is used in machine learning techniques to predict the body score of each animal individually, maximizing the return value.
- Project Leader
- Harsh hardware conditions on the field
- Worked on architecture and backend.
Enapark
Multiplatform app for parking payment, specially for malls.
A highly distributed system, with physical clients in half of the participating malls due to network restrictions on their internet policies. Capable of processing more than 5.000 payments per minute on the current version, the system has been stable since 2013.
With some malls being over 1.000 km away from our headquarters a lot of effort was made to keep our on-site hardware simple, durable and plug-and-play. Our efforts were rewarded as we never had issues that could not be solved remotely.
- Backend Development
- Architecture
- Mobile Team Leader
BH-TEC Talks
I currently co-coordinate the monthly meetings at Belo Horizonte Technology Park (BH-TEC). A free and open meeting where people from the park, and other companies come to discuss the latest techs and tendencies. All presentations are voluntary, made by our community for our community.
- 2+ years
- Free snacks =)
Portfolio optimization for open-pit mining assets acquisition
My master's degree work.
A Software capable of prioritizing the investments to be performed in an open pit mining operation, allowing an optimal portfolio considering budget limits.
I have worked on this project from conception to delivery and maintenance. A very enlightening experience as this was the project where I was able to integrate research and development in a whole new level. Both industry and academy walked hand in hand to achieve the intended result that was both profitable and pushed the boundaries of science by a little bit.
Being able to bind my work with my research was a very pleasing experience as it all converged to a single common goal. Having a research project backed by one of the world's biggest mining companies also helped the recognition and validation of my research by a significant ammount.
Curtida.Social
Have you ever seen a facebook post that claimed that, for each like it got, someone else would get some humanitarian help ?
Curtida.social (or social.like in a free translation) is just like that. Except it actually WORKS .
A platform where people or groups can propose any kind of social action and ask for money to perform such action. We search and find a company that also wants to see that social action happen and bring them together.
After they are brought together the social proposer must seek for popular engagement in social medias (facebook, twitter, youtube). The more engagement the campaign gets, the more money the backer company will contribute.
It's a win-win-win situation, the company gets both direct and indirect social engagement and marketing, while helping the society. The proposer gets to realize a dreamn and help people. And the society gets a better place by simpling hitting a like button.
- Worked on conception, backend and architecture
- Rewarding experience
- Brazil-only for now
Move
The guys at WayCarbon, (which worked a floor above us at ENACOM) had a problem: they had enviromental models made by developers, those models could chained together but the ones who knew how to chain the models were consultants, who had no idea how to program.
And now we come in: we developed a portal whith easy to use interface where consultants could drag and drop models together, clients could access the results in real time, and it all integrated with the developers versioning tools, and moved all the processing from the developers computer to a cloud environment streamlining the process and making it more eficient for everybody.
Although it seems simple, it demanded heavily on our cloud provider knowledge to keep bills low and processing time efficient as inputs could be many GB in size and processing times could be bigger than ten hours.
- Worked on all levels of the project, from conception to delivery
- Coordinated the entire team
- Responsible for architecture
- Responsible for training both the consultants and developers in the new tool
Fault Detector
My first project at ENACOM. Based on a academic research from another collaborator, this software was capable of using a brand new and more efficient approach to find the point of fault in long transmission lines.
This is a real problem in Brazil as transmission lines can be several hundred of kilometers long, so pinpointing the exact location where a problem happened (tree falling, fire, rupture, etc), specially in remote regions, can be a big time saver.
- Backend Development
- Architecture
Ugli or Duda
This will sound insane but bear with me:
I had a cat (Duda) and a Puppy (Ugli). Duda was very old and had some liver problemns thus needing some special food, which was extremelly expensive. She was small and didn't eat a lot so no huge deal.
Ugli on the other hand, although skinny, did eat a lot. Her food was not that expensive, so no big deal.
Duda's extremelly expensive food was ending too fast, and I got suspicious. As a godd computer vision student at the time, and inspired by a kaggle competition I did what any good nerd would do:
I took my Raspberry Pi (with a camera) and built a Computer vision algorithm that could diferentiate Ugli from Duda. I placed the camera pointing towards the expensive food and configured it to send me emails with pictures of the crime should Ugli ever try to eat it.
And confirming my suspicion Ugli was really eating the expensive food! Mystery solved!
Of course I could have moved the food to a place where Ugli wasn't able to reach in the first place, but where is the fun in that ?
- I had a lot of free time back there
- I submitted this as my final project in class and got an A+
Various Gaming Projects
I have a passion for gaming and video games in general. Due to this interest I have learned, on my own, how to develop games from scratch.
I have been educating myself in this area for quite some time. Not only in development but also in game marketing, game art, workflows and general game design. In this process I have learned and aquired several usefull skills out of what started as a simple hobby.
Most of my work is public on GitHub and I have several videos of my unfinished games in this youtube playlist. Although I have never published a complete game, I want to do so eventually.
- Every line of code and idea
Orange Rust Detection
A proof of concept project that never became a full fledged product, this piece of software aimed to identify diseases on sugar cane crops.
Drones would fly over the crops and aquire high quality images that would later be processed and used to identify potential diseases before they could spread.
The initial proof of concept was able to identify a single disease known as Orange Rust and also determine how advanced the disease was, although early results were promissing the projec was cancelled due to unforseen complications.
- Developed the entire POC
PoupaPark
Platform with the objective of integrating stores, sponsors, parking lots and clients through the concession of disocunts in parking prices for final users.
- Worked on Backend and architecture
- Mobile Team Leader
Work Work
ENACOM's internal control of working hours, developed from scratch.
Due to the awesome and quite unique working conditions at ENACOM, we had never found a suitable system for controlling projects and manpower. After trying several adaptations we simply decided to make our own, that would reflect the company's values and objectives.
- Conception
- Backend Development, Architecture
- Ever evolving software
Automatic Teeth Detection For Excavator Buckets
Computer vision software capable of identifying excavation shovel parts.
During the excavation process the existing 'teeth' in the shovel are subject to tearing and rupture. Detecting the state of such components is not only important for maintenance purposes but also for avoiding load contamination.
- Every single line of code
- Part of a bigger project
Enacoins
One of my biggest contributions to the workspace at ENACOM, enacoins are a currency that employs several game theory elements allied with many of the guidelines of healthy competition that I have learned studying gaming, to deliver a rewarding experience to the collaborators while motivating them.
Enacoins were propposed and planned in 2017 and started circulating in 2018. This currency is used to promote team awareness and bonding in general for most of the time, but can also be used to motivated short term behaviours when needed.
Being very well recieved by ENACOM collaborators, enacoins can be converted in parties, barbecues, surprise cakes, and a series of other interesting elements that change over time.
- Conception
- Implementation
- Continuous analysis of enacoins effects on workplace
Home Office and Productivity (Blog Post Colab)
I did this nice collaboration with the Belo Horizonte Technology Park (BH-TEC) blog, where I talked about Home Office and productivity.
I think Anna did a really good job in her text and captured the essence of our conversation. Give it a look - if you can read portuguese.
- Not my best picture
- That was my room at the time