I’m Fitbod’s Head of Data Science, Research & AI. Fitbod creates strength-training workouts that adapt to your body, recovery, and progress. I joined the team in 2021, after being a passionate Fitbod user for a few months.
At Fitbod I’m responsible for the algorithm that generates each workout, for analytics and A/B testing, for applied research, for patent applications, for hiring and developing data scientists, and for ensuring other teams get correct answers when they ask Claude and Codex data questions. I also participate in Fitbod’s overall business planning and help build the product roadmap.
Some things I’ve done at Fitbod:
Set up Fitbod’s A/B testing framework
I set up Fitbod’s A/B testing methodology and pipeline, enforced the same standards across all squads, and worked with my team and Product to choose appropriate KPIs and guardrail metrics. We went from 4 A/B tests in 2021 to 57 in the first half of 2026.
Sped up algo development by 9x
When I joined Fitbod the algorithm that generates the workouts ran on the client. Any algo changes had to go through the client teams, which slowed down development. I rewrote the entire algo in Python and migrated it server-side. That put the algo under my direct responsibility and enabled us to ship 9x more algo A/B tests per year.
As a bonus the iOS and Android algo are now the same (there were discrepancies before), we get to update it without users having to re-install the app, and server-side logging has made it much easier to reproduce algo output and investigate user complaints.
Increased exercise compliance by 42%
With the algo running server-side, my team and I were able to leverage our 10 years of workout data to personalize the workouts more. As a result, we’ve reduced the rate of replaced and skipped exercises by 42%.
Increased sets/reps/weights accuracy by 46%
By leveraging our 10 years of workout data, my team and I have made our capability estimation much better: users went from following our exact sets/reps/weights recommendations 30% of the time to doing it 44% of the time.
Sped up strength gains by 4x
In 2021 the average Fitbod user could bench press 6% more weight after doing the exercise ten times. In 2026 that’s 26%. That’s a direct result of more personalized workouts.
Did research with leading sports scientists
I reached out to professors Brad Schoenfeld and Andy Galpin and pitched a research collaboration. They said yes, brought along Dr. Federica Conti, and the four of us just co-authored a research paper that looks into the determinants of long-term strength training adherence - upcoming in Frontiers in Sports and Active Living.
Created a better 1-Rep-Max formula
Your 1-Rep-Max is the maximum weight you can lift for a given exercise in one set of one rep. It’s mostly a theoretical number, estimated from whatever sets/reps/weights you’ve actually lifted for that exercise.
1-Rep-Max estimation is noisy, based on formulas derived from small samples and only a few exercises. I mined Fitbod’s data - 303,494 near-failure sets from 14,966 users across 388 exercises spanning 16 muscle groups - to create a better formula. Pre-print here, currently under review.
Oversaw my team in projects that led to patents.
US12551760B2 (granted) and US20240299809A1 (pending).
Hired, mentored, and promoted data scientists to senior roles
The Observatory of Public Spending uses data science to find cases of corruption in federal spending. It is part of the Office of the Comptroller-Geral and it pioneered the use of machine learning in Brazil’s public sector. It also helped other government agencies leverage data science tools and inspired the creation of several data-driven units. I worked there as a senior data scientist and as the Observatory’s deputy head.
Some things I did at the Observatory:
Trained an ML model that helped the Comptroller-General of Brazil choose which municipalities to audit. Total auditable budget was US$ ~3.6 billion.
Trained an ML model that estimated the price of government-owned real estate properties, to help the government auction them.
Mined the database of covid economic relief to find cases of fraud.
Trained and mentored data scientists in other government agencies.
I have taught statistics and machine learning at different universities and research institutions - the University of Brasília, IESB, IDP, and IPEA -, both undergraduate and graduate courses. You can find some of my syllabi, slides, etc here.