atopos

atopos GmbH – enterprise AI consulting and ventures, Vienna. Out-of-place thinking. In-production systems.

atopos
01 – CLIENTS

Systems that stayed up.

Everything here is built inside the client's own boundary – their tenant, their network rules, a sealed DMZ where the data is not allowed to leave.

Borrower PDFs were opened one at a time and typed into a spreadsheet. Now they are read the moment they land in Outlook – text where there is text, OCR where there is not – the figures are lifted out, the lending ratios are computed from them, and the officer confirms instead of transcribing. Compliance status is a live matrix in the Teams app they already had open.

Outlook / MS Graph · OCR → LangGraph · Microsoft Teams

Read the case

Reporting ran on BW with Analysis for Office on top, and every month analysts rebuilt the same workbooks by hand. The numbers now come from Datasphere, an agent writes and runs the SQL, and the report is assembled, versioned and handed back as Excel, PDF or DOCX – inside the Teams app. The spreadsheets stayed; the rebuilding did not.

SAP Datasphere / HANA · LangGraph + DuckDB · Microsoft Teams

Read the case

When a customer said they called on Tuesday, twice, the answer depended on who you asked. Now it doesn't: every call lands in one queryable record you can cite, refreshed every six hours – and a second dealership runs on the same system, sealed off by credential.

TELUS Business Connect · Postgres · credential-isolated tenants

Read the case

02 – PARTNERS

Who we build with.

The client systems above were delivered with these companies and people.

03 – METHOD

The demo always works. So we skip it.

01
We build first

One slice that matters – and which slice that is, is a question about your organisation before it's a technical one. We build it against the real systems: your SAP, your tenant, your network rules. At our cost.

02
You decide after

You judge it running on your own data, not from a deck. Weeks, not months. If it doesn't hold up in your hands, nothing is owed and nothing is installed.

03
Then we stay

Go-live is the middle, not the end – the evals, the logs, and the inputs nobody predicted. That part decides whether it is still running a year from now.

Between engagements the systems keep working on themselves – knowledge bases that re-index, an org-memory graph that remembers why a decision was made, agents on our own servers running scheduled research and health-checks. You come back to something better than you left.

  • LangGraph
  • FastAPI
  • Python
  • Azure AI Foundry
  • Microsoft 365 + Teams
  • SAP Datasphere / BTP
  • PostgreSQL
  • DuckDB
  • Neo4j
  • Terraform
  • Hetzner
  • Dokploy
  • MCP
  • Claude Code
  • Modal
  • Next.js
THE BORING PART IS THE POINT
04 – JOINT VENTURES
Not everything we build has a client.
Own products and startup builds, on the same stack and the same servers as the client work. It's where we find out what a system does in year two – and the consulting gets the scar tissue for free.
buildrunreinvest
atopos
response < 48h
Wien / remote
© 2026 atopos GmbHpreviously polly consulting e.U.
ImprintPrivacy