SpaceFind

SpaceFind is a "Command+F" for the physical world. Using AR and a vector database, we index items in 3D space so you can ask "where is it?" and follow an instant AR path to your lost equipment.

Comment

Inspiration

It started with a universal frustration: the frantic hunt for a wallet and keys right when we needed to head out for the hackathon. We realized that while we can find a single word in a thousand-page document instantly using Ctrl+F, finding a physical object in a messy room still requires manual labor. We wanted to bridge this gap, bringing the power of digital search to the physical world to eliminate "search time" from our daily lives.

What it does

SpaceFind is a "Command+F" for reality. It is a spatial database with integrated AR that allows users to index physical items in real-time. As you move through a space, the app identifies objects and saves their 3D coordinates into a vector database. When you lose an item, you simply ask the app, "Where is my screwdriver?" or "Find my badge." SpaceFind retrieves the location and provides an augmented reality path—a visual breadcrumb trail—leading you directly to the lost item.

How we built it

SpaceFind is built at the intersection of Computer Vision, Spatial Computing, and Vector Databases.

  • The Index: We utilized Vision-Language Models (VLM) to process live camera feeds, identifying objects and converting them into high-dimensional embeddings.
  • The Memory: These embeddings are stored in a vector database, which allows for semantic searching. This means the system understands context, allowing for searches like "the heavy tool" or "the red pouch."
  • The Guidance: By integrating AR frameworks, we anchor items to specific 3D coordinates. The app then calculates the path from the user's current position to the item's stored location.

Challenges we ran into

The biggest hurdle was spatial persistence. Maintaining a digital marker's exact position as a user moves through different rooms required precise coordinate mapping and handling "drift" in the AR environment. We also had to optimize our vision pipeline to ensure that indexing happened in real-time without overwhelming the device’s hardware, balancing high-accuracy VLM processing with the need for low-latency feedback.

Accomplishments that we're proud of

We are incredibly proud of achieving real-time semantic retrieval. It was a major win to move beyond simple keyword tags and create a system that can find an object based on its visual description. Seeing the AR path accurately snap to a hidden object for the first time felt like watching science fiction become a functional tool.

What we learned

This project taught us that the future of productivity lies in Spatial Intelligence. We learned how to transform unstructured visual data into a searchable, structured database of reality. On a technical level, we gained deep experience in managing vector embeddings and optimizing on-device AI for mobile environments.

What's next for SpaceFind

We want to expand SpaceFind for enterprise and industrial use. Imagine a warehouse where new employees can find any tool among thousands of bins using AR, or a hospital where nurses can locate life-saving equipment instantly. We also plan to integrate "multi-user syncing," allowing a team to maintain a shared spatial catalog where if one person sees an item, everyone knows where it is.

Built With

  • android-speech-recognition-&-tts
  • androidx
  • backend
  • djl/sentencepiece-for-tokenization
  • kotlin-android-app-with-arcore
  • litert-+-litert-lm-on-qualcomm-npu-(qnn)
  • opengl-camera-preview
Share this project:

Updates