Inside the source layer: how SNAP grounds every training video in your own material
product
A deep look at the source layer — captures, extraction, linking, indexing — that makes one-sentence training video requests actually work.
The reason one sentence is enough to generate a training video isn't the model. It's what sits underneath: a source layer that captures your company's material once, then keeps it structured and current. Here's how each part works.
Capture: many formats, one destination
The source layer accepts what your team already produces — PDFs, slide decks, internal pages, screenshots of dashboards or UI, voice notes from subject-matter experts, meeting recordings. Nothing is reformatted for the tool; the tool adapts to the format. Each capture is timestamped and attributed.
Extraction: turning files into facts
Once a capture arrives, extraction pulls out entities (products, policies, roles, procedures), relationships between them, and the atomic claims each source makes. A 40-page compliance handbook becomes hundreds of linked facts, each traceable to a specific page and paragraph.
Linking: where the graph earns its keep
Extracted facts don't sit alone. They link to existing nodes: the new refund policy attaches to the customer-support cluster, the updated brand kit connects to every deck that referenced the old one. This is why a request for "onboarding for a new AE" can pull compensation, tooling, product, and process material in a single pass — the connections already exist.
Indexing: retrieval that respects structure
Retrieval reads the graph, not just raw text. When you write a one-sentence request, the system walks the graph from the topic you named outward, gathering related nodes with their citations. Plot then narrates from that gathered set and renders the finished video.
Why this changes the request pattern
Because the source layer is doing the heavy lifting, the request itself can stay short. You don't hand-select clips or paste in outlines. One sentence in, one finished video out — and the output stays grounded in material you already own.
For practical examples, see turning your org's knowledge into training videos on demand or the deeper walk-through of how captures become linked memory. Ready to try it? Start with SNAP.