About
NeuralPrint takes any stimulus, say a YouTube link, a video, an audio file, or text, and runs it through a real neural encoding model (TRIBE v2). The model predicts how 368 brain regions would respond over time, across vision, language, and sound. The result renders as an interactive 3D brain you can explore.
Key features
- Submit video, audio, text, or a YouTube URL
- Speech transcribed with Whisper large-v3
- TRIBE v2 predicts activation across 368 brain regions
- Interactive 3D brain (react-three-fiber)
- ROI scoring & ranked region interpretation
- AI-generated plain-language insights
Technologies
Next.jsThree.jsFastAPIModal GPUTRIBE v2
How it works
- 01You submit a stimulus: video, audio, text, or a YouTube URL.
- 02Modal runs it on an A100 GPU, and Whisper transcribes any speech.
- 03Per-modality encoders feed the TRIBE v2 model, which predicts fMRI responses.
- 04An interactive Three.js brain shows where and when activity happens.
Why it matters
A genuine multimodal deep-learning build: vision, language, and audio modelled together against measured brain responses.
ML research made interactive. The hard part here is the model and the inference pipeline, and that is exactly what was built.
Want something like this?
We'd scope it to your data and your workflow.