A RAG engine that answers from your own documents
Upload PDFs, Word files and spreadsheets, or import a website — the assistant answers from that content and nothing else.
- Year
- 2026
- Duration
- Ongoing
- Team
- In-house
- Sector
- AI / Retrieval-augmented generation
The problem
A language model on its own answers confidently from nothing, and one invented price or policy costs more trust than a hundred correct answers earn. The facts a business needs answered live in its brochures, price lists, policies and website, not in the model.
What we did
Documents are extracted, split into sections of around 900 characters, and indexed in Postgres with both full-text search and pgvector embeddings. Each question runs a hybrid search — keyword ranking fused with semantic similarity — so exact product names and loosely worded questions both find the right passage, including questions asked in Roman Urdu. Only the best-matching sections are given to the model, low-similarity matches are dropped, and a test-search screen shows exactly what the bot will read before a customer ever asks.
What you get
The retrieval engine behind every ChatDesk bot: answers grounded in the business's own material, kept current by re-uploading a file rather than retraining anything.
Tell us what is breaking.
Send the problem, not a polished brief. An engineer reads every enquiry and replies within two working days with a real technical opinion — including when the honest answer is that you do not need us.
- Reply from an engineer, not a salesperson
- Fixed scope and price before any build starts
- Your code, your cloud, your repo — from day one