RavoTeq, own product · 2026 to now · In development
RavoShield
Teams use AI on their internal data without their sensitive data leaving their control. One of six nominees for the Dutch Privacy Awards 2027 (Nederlandse Privacy Awards), chosen by an independent jury.
1 of 6nominees, Dutch Privacy Awards 2027
The problem
Organisations sit on data that could answer real questions, and most of it is too sensitive to paste into an AI tool. So they choose between a legal review for every question and not asking at all, while their people use public AI tools anyway.
Chat
Summarise for the team: Marieke Jansen, 52 years old, BSN 123456782, on insulin. Call +31 6 1234 5678.
Documentrefund.pdf
- Holder
- Jan de Vries
- IBAN
- NL91 ABNA 0417 1643 00
- Amount
- €1,240.00
Files3 files
PDFpatient_intake.pdf
XLSpayroll_q3.xlsx
DOCclient_contract.docx
Confidential risk 86 / 1008 values protected before sending
My role
CTO at RavoTeq, architecture and build
Where it stands
- 0
- outbound connections in the self-hosted mode
- 4
- major AI assistants covered: ChatGPT, Claude, Gemini, Copilot
Stack
- RAG
- MCP
- Self-hosted models
- Policy guardrails
- Audit trail
What changes
AI on sensitive data
WithoutA legal review per question, or the question never gets asked.
With RavoShieldTeams use ChatGPT, Claude, Gemini or Copilot, with sensitive data protected before it leaves.
Shadow AI
WithoutPublic AI tools used anyway, out of sight.
With RavoShieldPeople keep their tools. The organisation keeps its data.
The most sensitive data
WithoutCannot leave under any processing agreement.
With RavoShieldThe model runs inside the perimeter, with zero outbound connections.
Proof for the regulator
WithoutPrivacy claims stated, not shown.
With RavoShieldAn audit trail of what was protected and how each answer was reached.
How it works
At a high level. A detailed walkthrough is available on a call.
- 01Connect
- 02Protect
- 03Contain
- 04Govern
- 05Prove
Decision 01
Protect data before it leaves, instead of blocking AI tools.
People keep the tools they already use, and the organisation keeps its data.
Decision 02
A self-hosted mode with zero outbound connections.
Some data cannot leave under any processing agreement. For that data, the model comes to the data.
Decision 03
Guardrails set as policy, per team and per person.
A hospital ward and a marketing team do not need the same rules.
Decision 04
Every protection and every answer leaves an audit trail.
A privacy claim has to be provable to a regulator, not just stated.