Tutorial: Private AI Document Analysis

Watch Gemma 4 E2B run entirely in your browser — from model download to knowledge graph and Q&A. No upload, no cloud.

3 min · no audio · captions included

Step 1 — Load the model

Gemma 4 E2B downloads once to your browser cache and runs via WebGPU. No API key, no server.

Step 2 — Open a PDF

PDF.js extracts the text in the browser. The file never leaves your device.

Step 3 — Build the knowledge graph

The model extracts entities and relationships into a DuckDB-WASM knowledge graph stored in-browser.

Step 4 — Ask questions

Q&A retrieval grounds answers in the source text using the knowledge graph as an index.