Memory Box
This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle What I Built MemoryBox is a privacy-first screenshot search tool that helps students, developers, and researchers find images using natural-language queries. This web demo requires manual screenshot uploads and uses OCR, Gemma 3 4B, and local embeddings for visual understanding and semantic search. The actual…
MemoryBox is a privacy-first screenshot search tool designed to assist students, developers, and researchers in locating images through natural-language queries. This web-based demo necessitates manual screenshot uploads and employs OCR, Gemma 3 4B, and local embeddings for visual comprehension and semantic search. The envisioned future is a desktop application that, with user consent, accesses and indexes chosen screenshots and images, executing open-weight AI models locally. The intended business model involves a one-time desktop license.
MemoryBox operates as a browser-based demonstration, utilizing Tesseract.js for OCR and Ollama for local AI inference. Gemma 3 4B scrutinizes screenshots and generates visual descriptions, while the open-weight nomic-embed-text model generates embeddings for semantic search. A hybrid search system amalgamates keyword matching with vector similarity, preserving metadata and embeddings within IndexedDB.
All AI inference transpires locally via Ollama, devoid of cloud AI APIs. The ultimate objective is to transition this process into a desktop application with permission-based access to local screenshots.
The underlying idea behind open innovation is to facilitate privacy-first, local AI without reliance on paid cloud APIs. By utilizing open-weight Gemma and nomic-embed-text via Ollama, users can analyze and search their personal screenshots on their own devices, safeguarding sensitive data locally. Moreover, it empowers developers to scrutinize, replace, and enhance models, augmenting MemoryBox's adaptability and affordability.
The first user of MemoryBox was the reporter's sister, who utilized it to locate screenshots she couldn't find manually. Impressed, she paid the reporter ₹100 for the service - the reporter's initial revenue from the project. The second user was a friend who admired the concept and offered to share around $200 worth of API credits he won in a competition to aid the project's development. These early responses demonstrated that MemoryBox could address a genuine problem for individuals.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
