An AI lab,
reasoning from relationships.
Avra is a frontier AI lab building a predictive platform for enterprise decisions. We pre-train a Graph Foundation Model on the relational economy, then fine-tune inside each customer's workspace — for credit, fraud, growth, monitoring, and any task that is shaped like a graph underneath.
- Founded
- 2024
- Headquarters
- São Paulo
- Operating model
- Remote-first
- Team
- ~20 people
Built where the
relationships are.
Avra started from a simple observation. Every important enterprise decision — extending credit, catching fraud, predicting growth — depends on relationships that the source systems can see but the models cannot. The tables are rich. The flat features fed into the model are not.
We set out to change the shape of the input. Instead of rows and snapshots, we pre-train on a temporal knowledge graph of the economy: companies, individuals, and the events that connect them over time. The result is a Graph Foundation Model — one pre-trained, fine-tuned per customer, that improves the decision systems enterprises already run.
We are not trying to replace the credit bureau, the case management tool, or the marketing platform. We sit underneath them and make every decision they take a little sharper. The lab side of Avra publishes the research. The platform side ships it into production through shadow deployments, API and batch surfaces, and the governance enterprise teams already require.
Avra is small on purpose. Twenty people, deep stack, no contractors on the model path. Engineers, scientists, and operators who own the work from data ingest to live decision.
Model relationships over time.
Improve the decision systems enterprises already run.
Traditional systems reason from rows and snapshots. Avra reasons from relationships over time. We believe the next generation of enterprise prediction is graph-native — and that the companies who run on this substrate will compound faster than the ones who don't.
The two sides
of the company.
One came from enterprise credit. The other came from frontier AI. Avra exists because both sides decided the decision layer was where the work had to happen.
- Co-founder · CEO
Viviane Meister
A decade in enterprise credit before Avra. Built portfolios and risk functions across IFC, Itaú BBA, Santander, and HSBC, and spent four years as a partner at idwall. Avra is the company she wished existed when she was sitting on the other side of the table.
- Co-founder · CTO
Bruno Alano
Researcher at OpenAI in 2016, the youngest at the time. Co-founded and led engineering at Leadster; built AI teams at Isaac and Jusbrasil. Founding member of the Brazilian AI Association. At Avra, leads the lab and the platform that sits underneath it.
Twenty people, deep stack, real ownership.
Backed by frontier funds across two continents.
We chose investors who understand both sides of Avra — the frontier research and the enterprise deployment. They have backed the labs and the platforms we measure ourselves against.
Scout
Scout
We're hiring.
Come build the lab with us.
Avra is a small team of engineers, scientists, and operators shipping predictive systems into enterprise production. We're growing the lab and the platform underneath it, and we hire for depth and the discipline to ship.