FotoFindr

Inspiration

Modern camera rolls have thousands of photos but almost no meaningful organization. Finding a specific memory usually means endless scrolling. We wanted to make photo libraries searchable like the web, so you can find moments using natural language instead of manually browsing.

What it does

FotoFindr turns your camera roll into a fully searchable AI memory engine. Instead of scrolling, you can type queries like:

“Happy dog at the park”
“Pictures of the beach”

Our system analyzes every uploaded photo using computer vision to detect objects, faces, emotions, and scene context. The results are indexed and searchable using semantic AI search.

Users can also:

Detect low-value images
Generate AI narration describing photos and memories

How we built it

We built FotoFindr as a full-stack AI system.

Frontend

React Native with Expo for a fast mobile interface

Backend

FastAPI server handling uploads, metadata, and search

Vision & AI Pipeline

YOLO for object detection
DeepFace for face detection and emotion analysis
Gemini for semantic caption generation

Search System

Photos are converted into embeddings and stored in Snowflake
Natural language queries are embedded and matched using vector similarity search

Voice Narration

ElevenLabs generates spoken descriptions of photos and search results

Cleanup Photos

Detect photos that do not have a clear object that may be taken by accident will be marked for deletion

Together, this pipeline converts raw images into a structured, searchable knowledge graph of personal memories.

Challenges we ran into

One major challenge was combining multiple AI models into a fast pipeline. Image captioning, object detection, face analysis, and embedding generation all needed to run efficiently without slowing down the user experience.

Another challenge was semantic search accuracy. Translating natural language queries into meaningful filters required careful structuring of metadata and embeddings so results actually matched user intent.

We also had to balance powerful AI features with hackathon time constraints, focusing on the most impactful capabilities.

Accomplishments that we're proud of

We're proud that we built an end-to-end AI search engine for personal photos in a short time.

Automatically analyzing photos with multiple computer vision models
Building a working semantic search system
Generating AI voice narration of memories

Most importantly, the experience feels magical: you can type a sentence and instantly rediscover moments from your camera roll.

What we learned

This project taught us a lot about building AI-first applications. We learned how to:

Combine multiple computer vision models into a unified pipeline
Implement vector search for natural language queries
Design systems that transform unstructured data (images) into searchable knowledge
Build fast prototypes that still feel like real products

It also showed us how powerful AI becomes when it’s applied to personal data and memories.

What's next for FotoFindr

Future features could include:

Smart photo cleanup suggestions
Timeline-based storytelling of events
Automatic highlight reels of trips and milestones
Privacy-first on-device processing
Integration with existing photo libraries Our goal is to make personal photos not just stored, but truly understood and searchable.