Self-built product
Bastion
A private RAG system that runs entirely on a server you own, answering from your documents with the exact passage, page and section.
A complete product I designed, built and tested myself. Client work is in Case studies.

Some documents can’t leave your server
Many teams want to ask questions of their own documents but can’t send them to a public AI service. Bastion keeps the whole system, model included, on a server you own.
Answers from your documents, and says so when it can’t
Bastion answers from PDFs, Word files and web pages with the exact passage, page and section, and says so when the answer isn’t there. Text inside a document that tries to give the AI orders is flagged and treated as plain text.
People log in with admin, editor or member roles; rate limits and an audit log keep use in check.
Local model, one install script, backups that restore
The model runs locally with Ollama, so no data leaves the server and there are no per-message API fees. One install script sets up Docker, the firewall, nginx and SSL on an Ubuntu VPS or AWS EC2. Nightly backups restore with one command, and updates take a backup first and show rollback steps if a health check fails.
Tech stack
- Python
- FastAPI
- Qdrant
- Ollama
- Docker
- nginx
Screens from Bastion

When the answer isn’t in the documents, it says so.

Text that tries to give the AI orders, flagged and treated as plain text.

The audit log.

System health and backups.

A rate limit in action.
Want something like this for your business?
Tell me what you need it to do. I’ll reply within 24 hours on weekdays with questions or a suggested first step.
Start with the $150 scoping session, credited toward the build.