We are a collective of data scientists and data-driven social scientists deploying machine learning, causal inference, and geospatial analytics to solve the hardest problems in developing economies.
We are economists, statisticians, and machine learning engineers who have spent decades inside the data systems of Latin America and the Global South. We don't just build models — we understand the surveys, the censuses, the policy constraints, and the people behind the numbers.
Most AI firms sell technology looking for a problem. We start with the problem — whether it's a government trying to target social programs or a bank trying to score credit for the unbanked — and engineer the minimum viable intelligence to solve it. Our stack is open-source, our methods are peer-reviewed, and our outputs are reproducible.
Where Palantir charges millions for opaque platforms, we deliver transparent, auditable analytics at a fraction of the cost — because the governments and organizations that need this most are precisely those that can least afford legacy enterprise pricing.
PhDs in economics and public policy who build production ML pipelines. Not data engineers guessing at social science.
R, Python, PostGIS, INLA, Stan. No vendor lock-in. Every model is reproducible, every result auditable.
Born in Latin America. We know the data ecosystems, institutional constraints, and languages — because we've lived them.
No surveillance. No individual tracking. Aggregate intelligence that protects the communities it serves.
End-to-end analytical capabilities for governments and enterprises — from raw data to actionable intelligence.
Granular, municipality-level estimates of income, purchasing power, and deprivation — the ground truth that drives smarter decisions. Governments use it to allocate resources; banks use it for credit risk; retailers and telecoms use it for market entry, pricing strategy, and distribution planning. Our three-layer methodology (DRF + BYM2 + Optimal Transport) delivers full uncertainty quantification for every small area across LAC.
From national vote-share forecasting to precinct-level swing prediction. We build ML models that integrate census demographics, historical electoral results, economic indicators, and sentiment signals to forecast outcomes and identify persuadable segments. Used by campaigns, political consultancies, media organizations, and risk analysts across Latin America.
The discipline of dividing populations into actionable groups — whether customers, voters, or citizens. We apply unsupervised learning, latent class analysis, and geo-demographic profiling to reveal hidden structure in your data. Identify high-value micro-segments, map voter coalitions, or build vulnerability typologies — all grounded in ground-truth socioeconomic intelligence unique to LAC markets.
AI-driven pricing that adapts to demand elasticity, competitor signals, and local purchasing power. We build pricing engines for telecom, retail, insurance, and financial services — calibrated to emerging-market realities where willingness-to-pay varies dramatically across geographies and income levels.
Alternative credit models for the 1.4 billion adults with no formal financial history. We combine predicted welfare, asset indices, utility payments, mobile data, and geographic risk to build credit scores that enable microfinance, insurance, and digital lending — while managing portfolio risk at scale.
Know what your customers will need before they do. We build propensity models, next-best-action engines, and geo-temporal demand forecasts for FMCG, pharma, telecom, and retail. Optimize campaign targeting, inventory allocation, and distribution routes — even in data-sparse emerging markets.
Where to open the next branch, warehouse, or tower. We fuse satellite imagery, nightlights, road networks, and foot-traffic proxies with economic data to power site selection, trade-area analysis, and competitive intelligence. The same models that map poverty also map purchasing power, market density, and infrastructure gaps — giving firms a ground-truth picture of opportunity at any scale.
Natural language interfaces that let executives, analysts, and field teams query complex data in plain Spanish, Portuguese, or English. "What's our market share in the Southern region?" "Which stores underperform vs. their catchment potential?" Answers in seconds, powered by RAG over your integrated data.
Turn-key platforms for governments and enterprises operating in emerging markets.
Poverty Intelligence Platform
An end-to-end SaaS platform for small area estimation. Governments upload census and survey microdata; Atlas returns ward-level poverty maps, vulnerability indices, and targeting recommendations — with full uncertainty quantification and interactive visualization.
Social Program Targeting Engine
Intelligent beneficiary selection that goes beyond proxy means testing. Focaliza combines predicted welfare distributions with geographic vulnerability scores to minimize inclusion and exclusion errors — ensuring transfers reach the people who need them most.
AI Decision Assistant
A multilingual conversational interface that sits on top of any integrated data system. Executives, analysts, and field managers ask questions in plain language and receive instant, sourced answers with visualizations. No SQL. No dashboards. Just answers.
Market Intelligence Platform
Real-time market intelligence for firms operating in emerging economies. Pulso integrates satellite data, transaction signals, mobility patterns, and census-derived purchasing power to deliver hyper-local demand forecasts, competitive landscape maps, and expansion opportunity scores.
Our core methodology stacks three complementary AI/ML layers, each correcting the weaknesses of the one before it.
We train a DRF on household survey data to learn the full conditional distribution of welfare — not just the mean. Applied to census records, this produces household-level probability distributions of income, poverty status, and deprivation. Unlike traditional regression, DRF captures non-linearities, interactions, and the entire shape of the welfare distribution at each point.
Input: Survey microdata (12K+ households) + Census (3.5M+ households)Output: HH-level P(poor), E[welfare], conditional variance
Raw DRF estimates are noisy in small municipalities. We apply a Bayesian BYM2 model (via INLA) that borrows strength from neighboring areas through a spatial adjacency graph. This smooths erratic estimates in data-sparse areas while preserving genuine geographic variation — the gold standard in disease mapping and small area statistics.
Input: Municipal DRF estimates + adjacency graphOutput: Spatially smoothed poverty rates with credible intervals
The final layer calibrates the entire predicted welfare distribution to match known survey benchmarks at the regional level. Using quantile-based optimal transport with cross-validated regularization, we ensure that aggregate statistics (department-level poverty rates) match official survey estimates exactly — while preserving the granular within-area variation that makes small area estimation valuable.
Input: HH-level DRF predictions + regional survey benchmarksOutput: Calibrated welfare distributions, raked to official statistics
Municipality-level poverty, extreme poverty, and poverty gap estimates for all 340 municipalities using Census 2024 and Household Survey 2024. Three-layer pipeline with full uncertainty quantification. Validated against official department-level statistics with 90% CI coverage.
Geo-granular demand modeling for a major telecom operator. Predicted purchasing power and connectivity gaps across 340+ municipalities to prioritize tower deployment, optimize ARPU-based pricing tiers, and build micro-segments for targeted plan offers — increasing rural subscriber acquisition by identifying high-potential underserved zones.
Built ML credit models for a microfinance institution serving the unbanked. Combined predicted welfare, asset indices, utility payment patterns, and geographic risk to score 500K+ applicants with no formal credit history. Reduced default rates while expanding the eligible borrower pool by 3x.
Site selection engine for a retail chain entering secondary cities. Fused satellite-derived foot traffic, census purchasing power, and competitive proximity data to rank 200+ candidate locations. Model predicted revenue within 12% of actuals in the first cohort of new openings.
Geospatial risk models for an insurer launching agricultural and health micro-insurance products. Combined climate data, crop yield predictions, and socioeconomic indicators to build location-specific pricing that balances affordability with actuarial soundness across diverse risk zones.
Built geo-temporal demand models for a consumer goods distributor. Predicted weekly product demand at the ward level using economic indicators, seasonality, and mobility data — optimizing delivery routes and reducing stockouts in peri-urban and rural distribution networks by 25%.
A mayor shouldn't need to understand distributional random forests. They need to know where to send the ambulance. We build intelligence that disappears into decisions.
Every estimate comes with code, data lineage, and uncertainty bounds. If you can't reproduce it, it's not science — it's marketing.
A perfectly specified logit with the right variables beats a billion-parameter model trained on the wrong data. We invest in understanding the problem before we touch the keyboard.
The communities we map are often the most exposed to surveillance harm. We work with aggregates, spatial smoothing, and differential privacy — never individual targeting.
Not watered-down versions. Not last decade's technology. The same rigor applied to genome sequencing and autonomous vehicles — applied to poverty, nutrition, and education.
Whether you're a government ministry mapping poverty, an NGO targeting interventions, or a firm entering underserved markets — we'd like to hear from you.