Back to case studies
Cross-industryRecent delivery
Semantic search & AI assistant layer
Retrieval-augmented search and assistants over quotes, job cards, notes and correspondence — grounded in each client's own records and permissions.
SupabasepgvectorFileMaker Data API
The starting point
Twenty years of quotes, job cards, notes and correspondence already held the answers, but finding them meant knowing the exact wording the original author used, or asking a colleague who might have retired.
What we built
- Embeddings generated and stored in Postgres with pgvector as records change
- Retrieval-augmented answers assembled from the client's own records
- Citations that link every answer back to its source record in FileMaker
- Permission-aware querying, so users only ever see what they already could
What it delivers
- Vector embeddings synced from the operational database
- Answers with citations back to the source record
- Permissions honoured at query time
The outcome
Staff search by meaning instead of keyword, answers arrive with the source record attached, and the assistant honours the same permissions as the system behind it.
Typical modules
Vector syncSemantic searchRAG answersCitationsPermissions
FAQs
Questions about projects like this
Your system next
Bring us the process that looks like this one.
A 30-minute discovery call is usually enough to see how closely your operation matches this pattern — and what it would take to adapt it.