We built a second brain that fact-checks itself, running Gemma 4 locally, in one hackathon day

Description and use

Connectore links notes conceptually, visualizes them in a graph, and verifies AI story fidelity locally using Gemma 4.

Problem

Notes live in silos and AI summaries often drop critical facts, making it hard to see connections and ensure accuracy.

Outcome

Connectore links conceptually related notes across subjects, provides a visual graph of relations, streams concise stories, and audits those stories for factual fidelity—all running locally on a laptop with Gemma 4 and Ollama.

Workflow

• Write a note in the editor
• Gemma tags the note with summary, entities, and abstract patterns
• SQL shortlist finds up to 8 older notes sharing an entity or pattern
• Gemma judges each candidate for relation, strength, and reason
• Links with strength ≥4 are kept and displayed in a force‑graph
• User can drag one note onto another to judge a single pair
• Press "Tell the story" to stream a narrative tying the thread together
• The verifier runs a deterministic cue diff and a fact quiz to flag missing or altered facts

Evidence

Connectore is a local‑first second brain. Gemma 4 links your notes across subjects, tells you the story behind them, and then checks its own story against your notes for facts it dropped.

Source

Tag. Gemma 4 returns JSON with a summary, entities (RBI, TCP), and abstract patterns like feedback loop, trade‑off or oscillation.

Source

The verifier on a deliberately sabotaged story flagged all four cues and marked 3 facts as possible mismatches and 1 as needs review, in about 7.7 s.

Source

Pass num_ctx on every single call. Ollama's default context window depends on your VRAM and can be as small as 4K on a laptop.

Source