This is a submission for the Gemma 4 Challenge: Build with Gemma 4

Brain Dump — AI That Connects Your Knowledge

What I Built

Brain Dump is a second brain that turns scattered personal data into a searchable memory layer.

Users can capture notes, URLs, PDFs, screenshots, audio, video, and project files. The system extracts content, transcribes media with Whisper, builds embeddings, and enables semantic search across the entire knowledge base.

Instead of remembering filenames or folders, users can search naturally and retrieve the right context instantly.

Key Features

• Semantic search across mixed-format content


• Whisper-powered transcription


• AI-generated summaries and insights


• Memory connections + lightweight knowledge graph


• SQLite-based MVP with production-ready migration path to Postgres + pgvector

Demo

Video walkthrough coming soon.

Code

Repository:


https://github.com/siddqamar/brain-dump

How I Used Gemma 4

Brain Dump uses gemma-4-26b-a4b-it through Google AI Studio API for contextual summarization and insight generation.

I chose the 26B A4B model because it provided a strong balance between reasoning quality, speed, and practical deployment for a hackathon-scale product.

Gemma powers:

• Memory summarization


• Context extraction


• Insight generation


• Relationship discovery between captured knowledge

The goal was to combine local knowledge management with modern AI retrieval workflows in a lightweight and developer-friendly MVP.