Audit and deployment proposal. The platform was developed by another team.
The starting point
The Spanish Chamber of Commerce was evaluating running Atolónica, a platform that generates risk, impact and remediation matrices from regulations, without sending data to third parties.
The decision
Run the language model, embeddings and OCR on the server itself, with the data containers on an internal network with no internet access.
What we did
- Technical audit of backend, frontend, network and AI
- Target architecture with 100% local AI in the European Union
- Change and deployment plan, with its quote
Results
- 14 findings the previous analysis had missed
- With the proposed architecture, data cannot leave even if a URL is misconfigured
How quality was ensured
Critical findings were verified manually. Processing times were presented as estimates, to be validated in a one-month trial.
Technology
- FastAPI
- Celery
- PostgreSQL
- pgvector
- Ollama
- Angular