# PixaFind

> AI-powered semantic file search that runs entirely offline — find documents, images, and code by meaning, not filename. Privacy-first, no cloud, no accounts.

Canonical page: https://www.ishannaik.com/projects/pixafind

_Search your files by meaning, fully offline_

- Category: AI/ML Development
- Client: PixaLabs
- Role: Founder & Engineer
- Year: 2026
- Duration: 3 months
- Live: https://pixafind.com

## Overview

PixaFind is a desktop application that brings semantic search to your local files. Instead of matching exact keywords, it understands meaning — searching for "vacation policy" surfaces the right document even if those words never appear in it. Everything runs on your own machine, with verified zero network calls.

It indexes 76+ file formats — PDF, DOCX, PPTX, EPUB, emails, source code across 30+ languages, and images. Image search works through text-to-image matching: type "sunset at the beach" and it finds the visual matches in your library. Local vision and embedding models power the whole pipeline.

The product ships as a download-and-run app for Windows with no sign-up, API keys, or accounts. Background indexing stays out of the way, searches return in under 100ms, and the AI models run locally with optional GPU acceleration or a CPU-only fallback. The core learning notebooks are open source and the app is source-available.

## Features

- **Semantic Search**: Finds files by meaning across 76+ formats — PDF, DOCX, PPTX, EPUB, emails, and more
- **Image Search**: Text-to-image matching: describe what you remember and find the visual
- **Code Search**: Semantic search across source code in 30+ programming languages
- **100% Offline**: All AI models run locally — verified zero network calls, fully private
- **Sub-100ms Results**: Instant searches with non-intrusive background indexing
- **No Setup**: Download and run — no sign-ups, API keys, or cloud accounts

## Technologies

Python, Local Embeddings, Vector Search, Computer Vision, Astro, TypeScript

## Process

- **Research** (2 weeks): Evaluated local embedding and vision models for accuracy vs. footprint on consumer hardware
- **Indexing Engine** (4 weeks): Built the multi-format extraction + embedding pipeline with an on-disk vector index
- **Search & UX** (3 weeks): Implemented sub-100ms semantic + image search and the desktop interface
- **Packaging & Pricing** (2 weeks): Shipped the download-and-run app with auto-updates and a one-time purchase model

## Related projects

- [Manufacturing Defect Detection](https://www.ishannaik.com/projects/defect-detection.md)
- [PixaShield](https://www.ishannaik.com/projects/pixashield.md)