Receipts, always.
Every claim ships with the artifact behind it. Every decision is hash-chained. If we can't show you the lineage, we don't ship the feature.
AVA Research is a small team of engineers and researchers based in Bucharest. We build the data infrastructure regulated industries can actually defend — five products on one engine, every output carrying its own audit trail.
Foundation models have transformed text and images. Tabular data is the next surface — and the gating constraint isn't capability, it's trust.
We started AVA Research in 2023 to solve a narrow problem at the intersection of two unforgiving industries: the messy spreadsheets data engineers spend their weekends fighting, and the audit chains regulators expect them to produce on demand. The tools we needed didn't exist. So we built them.
Three years later, that internal toolkit has become DAVA — five products on one engine, sold à la carte to teams who'd rather buy the parts than build them. Customers in healthcare, finance, and the public sector use us where the audit trail matters more than the demo.
The five products share one engine — same auth, same audit, same policies — so the second product you wire up costs a fraction of the first. The audit chain is built in, not bolted on. Self-host is a first-class deployment, not an enterprise upsell. We sign a DPA before processing personal data, our application servers and database run in Frankfurt, and we'll show you the source code under NDA.
If that's the kind of company you want to work with, talk to us.
These aren't aspirations on a careers page. They're the rules we argue about in pull request reviews.
Every claim ships with the artifact behind it. Every decision is hash-chained. If we can't show you the lineage, we don't ship the feature.
If your data is sensitive enough that you're reading this, self-host is your default. Docker bundle, JWT licensed, air-gap supported. Same product, same updates.
If we can't move you from “nothing” to “working sandbox” in five days, we're not the right vendor. Most pilots run end-to-end in three.
Application servers run on Fly.io in Frankfurt and Postgres on Neon in eu-central-1. Object storage is on Cloudflare R2, which is not pinned to an EU jurisdiction, and AI features call Anthropic's API, which is not EU-pinned either. The trust page lists every sub-processor.
Reasonable people can disagree. We've found these to hold up in pilot work and a few painful retrospectives.
Postgres, FastAPI, S3, Redis. We chose interesting tech only at the layers where the bet pays off — model orchestration, audit chain, sandboxing — and stayed boring everywhere else.
Then ship the next. Most of our products started as a single endpoint we could demo on a 30-min call. The product surface grew because pilot customers asked for it, not because the roadmap said so.
Engineers are on customer calls. There's no specialist class translating between product and engineering — the people writing the code hear what's broken from the people using it.
Audit logs, retention TTLs, evidence packs, GDPR Article 17 workflows, two-factor sign-in. None of these will get a launch tweet. They will get us renewals.
We turn down deals that would force us to ship faster than our test suite. We've sized the team to the volume of pilots we can run honestly. If you ask us how big we'll be in three years, we don't know — and we think that's fine.
We hire on demonstrated work, not interview gymnastics. If you've shipped something we'd respect, send it to work@avaresearch.ai. We read everything.
We don't want the biggest company. We want the most defensible one.
Bring us your hardest dataset. We'll clean it, audit it, and give you back a file you can ship — plus an evidence pack for your auditor. Most pilots run in under a week.