Member of Technical Staff, Research — Relational Foundation Models
Confirmed on the employer's own hiring board on Nov 6, 2026. First seen by Alion on Apr 20, 2026.
About Avra
Foundation models learned to read text, images, and code. The data that runs the economy hasn't had its moment yet. It's structured: customers, accounts, transactions, and claims, stored in relational databases, connected as graphs, and moving through time. Every credit approval, fraud block, and retention offer is a prediction over that data. Today each one is still a bespoke model, rebuilt by hand for every task, schema, and company.
We're building foundation models for structured data, such as relational databases, graphs, and time series. Take a relational database: one pretrained model should read any company's database, understand its schema from the names of its tables and columns, and adapt to a new prediction task without a feature-engineering project or a model trained from scratch.
Avra exists to build those models and the platform that turns their predictions into decisions running in production, in our cloud or inside an enterprise's own infrastructure. We're an AI research lab with our home in São Paulo, backed by Sequoia and Andreessen Horowitz, and we ship with enterprise design partners in banking, payments, telecom, and commerce, including Mercado Livre, Inter, PicPay, Banco BV, and Vivo.
About the team
We are a small group of researchers, engineers, and company builders, with our home in São Paulo and a team spread across San Francisco, Latin America, and Europe. We hire at all levels. The scope you take on comes from what you've shown, not from a title.
Our researchers have published widely at venues like NeurIPS and ICLR. You can find our work on our research page.
Your mission
Help build a foundation model that understands any company's relational data and predicts well enough to run its decisions, better than the hand-built models companies rely on today.
What you'll work on
Adapting to new tasks. How a pretrained model learns a task it has never seen, from examples in its context or through other mechanisms, what information it needs (which rows, how much history, which neighbors), and why it fails when it fails.
Pretraining from scratch. Architectures, objectives, and scaling for transformers over multi-table, temporal data.
Pretraining data. Which corpora, real and synthetic, teach a model to generalize across schemas, domains, and languages.
Time and leakage. Predictions that only see what was knowable at decision time. That's the difference between a benchmark number and a model a bank can trust.
Evaluation. Relational and tabular benchmark suites, strong baselines (gradient boosting, tabular foundation models, task-specific relational models), and protocols that keep comparisons honest.
Post-training. Fine-tuning and other ways to specialize the pretrained model to a customer's data and task, and knowing when each one is worth it.
What we hope you'll bring
Excellent fundamentals and implementation skills: you train models in PyTorch yourself, including multi-GPU runs you've debugged
Research taste: you can read a paper, find the idea that matters, and design the cheapest experiment that could prove it wrong
Methodological rigor: temporal splits, ablations, multiple seeds, strong baselines, and honest error analysis
Experience with structured data, such as tabular or relational learning, recommender systems, time series, or any domain where gradient boosting is hard to beat
Clear written English
Bonus points if you have
Worked on tabular or relational foundation models, in-context learning, or meta-learning
Pretrained transformers at scale, or studied their scaling behavior
Written Rust, C++, CUDA, or Triton, or made a data loader or sampler substantially faster
Shipped a model to production and lived with how it behaved afterwards
A record of strong research, whether in papers or in open-source work
We welcome academic researchers and researchers with unconventional backgrounds. A degree is not a requirement.
How we interview
Intro call (30 min). You, what you want to work on, and what we're building.
Deep dive (60 min). A project you're proud of, taken as deep as you want to go.
Research talk (60 min). You present past work to the team and take questions.
Founders (45 min).
References.
Logistics
Location: Remote. We hire in most countries through Deel. The team works across San Francisco, Brazil, Uruguay, and Europe, and we get together in person regularly, in São Paulo and elsewhere.
Compensation: Competitive salary and equity, depending on background and level.
Compute: GPUs for pretraining-scale experiments, plus a research budget.
Benefits: Flexible time off, health coverage, extended parental leave, and travel to the conferences where you publish.
We're actively hiring for this role, and the research team reads every application. If we don't move forward this time, please wait six months before you apply again.