About
I’m Gonzalo Barrera Borla, a statistician and machine-learning engineer in Buenos Aires. My models have priced programmatic ad auctions (Jampp) and NASCAR fan-club subscriptions, sized correlated parlays (joint Kelly) for a horse-race betting syndicate, flagged crypto-wallet fraud (Muun), and matched people looking for love (Sitch). I really like problems where statistics meets the P&L: risk, matching, pricing, optimization.
This blog is where I write things down at the length they need, mostly in Spanish. Corrections and questions are welcome by email.
Experience
- Sitch2025–First data hire at a matchmaking startup that replaces swiping with curated candidate lists. Replaced LLM-only affinity scoring with chat-NLP extraction, gradient-boosted match probabilities and greedy b-matching: from over 10K USD a month in OpenAI fees to a few bare-metal VMs, holding match quality while growing from 3K to 30K MAU.
- Muun Wallet2024–25Data scientist on the anti-fraud team of a self-custodial Bitcoin and Lightning wallet. Real-time incongruity detection plus on-chain and Lightning contracts took micro-fraud of about 2 BTC a month (over 200K USD a month then) to near zero.
- Move 372023–24Head of data science (team of 4) for a pari-mutuel horse-racing syndicate across LATAM and the US: joint Kelly sizing over correlated wagers, estimated joint distributions for exotic bets. Volume more than doubled at stable margins.
- Jampp2021–22Engineering manager, data science. Led the move from a bespoke online-SGD logistic regression to batch LightGBM, unlocking per-client, per-geo models; wrote the team’s career ladder and grew it to 12 people.
- Jampp2019–20Senior data scientist on the conversion-rate models (win rate, CTR, CVR) behind a real-time bidder scoring 1M+ auctions per second. Co-developed the income-optimization engine that replaced 100+ per-client bidding problems with one global objective; annualized revenue grew about 4×, from about USD 15M to 60M+.
- Jefatura de Gabinete de Ministros2016–19Built the data model and ETLs of a BI system integrating Argentina’s national public administration: payroll for about 400K civil and military employees and the roughly USD 156B national budget with its monthly execution.
Work & talks
- Fermat distances in kernel density classifiers (MA thesis, defense pending): density-based geodesic distances on Riemannian manifolds for nonparametric classification, with a
scikit-learn-compatible library. Blog series, code, thesis. Instituto de Cálculo, FCEN-UBA (2026). - Optimal bidding: a dual approach: the income-optimization engine behind globally optimal RTB bidding at Jampp. AdKDD and PyData AR (2019).
- Machine Learning in Graphs: a semester course on learning and optimization on graph-structured data, fully open. Materials and video lectures. With Martín Elías Costa, FCEN-UBA (2021).
- Si nos organizamos, nos enamoramos todos: embedding learned random estimates in linear optimization, applied to finding love without swiping. Notebook and slides. Sitch and Universidad de San Andrés, MCD (2026).
- The limits of electoral predictability (BA thesis): Monte Carlo simulations of election-night partial tallies for Buenos Aires legislative elections, from open data. Code, thesis. FCE-UBA (2014).
- pydatajson: CKAN open-data metadata library for Argentina’s national data portal, featured in the IADB’s “Code for Development” program.
Teaching
- Instituto de Cálculo, UBA2023–25Teaching Assistant: “Intro to Statistics and Data Science” and “Data Laboratory”; materials at fcen-amateur.
- FCEyN, UBA2021External professor, “Machine Learning in Graphs”.
- Mathematical olympiad coach2012–17
Education
- MA in Mathematical Statistics, UBA — coursework completed, thesis defense pending2017–Instituto de Cálculo; GPA 10.0.
- BA in Economics, UBA2009–14Facultad de Ciencias Económicas; magna cum laude, class valedictorian.
Elsewhere
GitHub · LinkedIn · gonzalobb@gmail.com
Todos a bordo del tototren! — this site, 2014–2026