Compass · Content Engine

Your knowledge, not the model's guesses

The DataStore ingests your PDFs, URLs and entire domains — with OCR and audio transcription — into a Weaviate vector store, so every AI lesson, roleplay and grade is grounded in your own material.

The grounding problem

Generic AI invents. Grounded AI cites.

An off-the-shelf model trained on the open internet will happily make up your refund policy, your product specs or your compliance rules. That's a liability in training, where a confidently wrong answer becomes a confidently wrong employee.

Why a buyer cares: retrieval-augmented generation (RAG) anchors every answer to your approved documents — dramatically reducing hallucination and keeping training on-message and defensible.

DataStore · knowledge ingestion
Summer-2026-brochure.pdf
PDF · 212 pages · OCR applied
Embedded
tui.co.uk (tour operator site)
Domain crawl · 148 pages
Embedded
ABTA-refund-policy.pdf
PDF · indexing…
Processing
Weaviate vector store 12,480 chunks indexed · RAG-ready
The DataStore pipeline

From raw source to grounded answer

Ingest almost anything, and Compass turns it into searchable, citable knowledge.

Ingest

Upload PDFs, point at URLs, or crawl a whole domain — plus OCR for scans and transcription for audio.

Chunk & embed

Content is split and embedded into a Weaviate vector store so meaning, not just keywords, is searchable.

Retrieve

At question time the engine pulls the most relevant passages from your documents as context.

Generate, grounded

The model answers from that retrieved context — so lessons, roleplay and grading stay true to your source.

Feed it anything

OCR and transcription mean nothing gets left out

Real corporate knowledge isn't tidy. It's scanned manuals, a recorded webinar, a product page, a 200-page policy PDF. Compass's DataStore handles them all — OCR reads scanned images, audio transcription turns recordings into text, and domain crawling pulls in a whole site.

Why a buyer cares: you don't have to re-author your knowledge base to use AI. Point Compass at what you already have, and it becomes training material.

Compass content sources
How you stay in control

Predictable AI spend, no vendor lock-in

Budget with confidence and pick the best price/quality fit — run the right model for the job and see exactly what every interaction costs.

Switchable providers

Choose between OpenAI, Gemini and Azure per use case. Buyers care: avoid lock-in and pick the best price/quality fit.

Token & cost accounting

Every interaction logs token usage and cost. Buyers care: predictable, transparent AI spend you can budget and bill.

Reduced hallucination

RAG grounding keeps answers tied to your docs. Buyers care: training you can trust and defend in an audit.

Grounded AI you can stand behind

Your documents in, accurate training out — with the cost of every interaction accounted for.

Book a demo
PDF
URL & domain ingest
RAG
Weaviate vectors
3
Model providers
OCR
Audio transcription
FAQ

Content engine questions

What does RAG actually change for our training?

Instead of the AI answering from its general training data, it first retrieves the most relevant passages from your uploaded documents and answers from those. The result is accurate, on-policy responses — and far fewer invented "facts" reaching your learners.

Can it handle scanned documents and recordings?

Yes. OCR extracts text from scanned PDFs and images, and audio transcription converts recordings to searchable text before they're embedded into the vector store.

Are we tied to one AI provider?

No. Model configuration lets you switch between OpenAI, Gemini and Azure, with token and cost accounting per interaction so you stay in control of both quality and spend.

Ground your AI in your own knowledge

Book a demo and we'll ingest a sample of your documents so you can see grounded answers in action.