Member of Technical Staff, Research - Relational Foundation Models

About the role

Avra is a frontier AI lab building relational foundation models from the ground up, with customers such as Mercado Libre, Telefonica, Banco Inter and Getnet. The team is small, based in São Paulo and spread across San Francisco, Latin America and Europe, and hires at all levels. This remote research role helps build a foundation model that understands any company's relational data and predicts well enough to run business decisions.

Key responsibilities

• Adaptation to new tasks: study how a pretrained model learns a task it has never seen, what information it needs and why it fails.
• Pretraining from scratch: architectures, objectives and scaling for transformers over multi-table, temporal data.
• Pretraining data: corpora, real and synthetic, that teach generalization across schemas, domains and languages.
• Time and leakage: ensure predictions only see what was knowable at decision time.
• Evaluation: relational and tabular benchmarks, strong baselines and honest comparison protocols.
• Post-training: fine-tuning and other specialization methods, and when each is worth it.

Required profile

• Excellent fundamentals and implementation skills, training models in PyTorch including multi-GPU runs you have debugged.
• Research taste: reading papers, finding the core idea and designing the cheapest experiment that could disprove it.
• 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; a degree is not required and academic researchers and those with unconventional backgrounds are welcomed.

Required skills

• PyTorch
• Bonus: tabular or relational foundation models, in-context learning or meta-learning
• Bonus: pretrained transformers at scale
• Bonus: Rust, C++, CUDA or Triton, or making a data loader or sampler faster
• Bonus: shipping a model to production

What we offer

• Competitive salary and equity, depending on background and level.
• GPUs for pretraining-scale experiments and a research budget.
• Flexible time off, health coverage, extended parental leave and travel to conferences.

The process includes an intro call, a deep dive, a research talk and founder references. Applications are reviewed by the research team; if there is no move forward, candidates are asked to wait six months before applying again.