NeuraFlow

Inspirationa

We were inspired by a simple question: why do some short-form videos go viral while others don’t, even when the ideas are similar? Moreover, two of our team members have created TikToks in the past to advertise SaaS products, so we felt that answering this question would be directly applicable to our lives and those of many of our friends.

Instead of relying only on vanity metrics after posting, we wanted creators to understand audience engagement before publishing by using brain-activity-informed signals and actionable creative guidance. There are tons of online courses and guides on how to attain user attention, but we wanted to find something objective.

What it does

Our platform analyzes videos to estimate engagement-related brain activity patterns across dimensions like attention, visual imagery, emotion, and overall engagement. It utilizes TribeV2, a model trained on over 500 fMRI scans to predict human brain responses to differing stimuli. It compares user content against a curated library of already-viral videos, then uses an AI chat assistant to explain findings in plain language and suggest specific improvements (hooks, pacing, emotional arc, scene structure, and A/B test ideas).

How we built it

We built a full-stack system with: FastAPI backend for analysis endpoints, upload flows, timeline/spike extraction, and similarity search Postgres/Neon for storing predictions, embeddings, timeline data, and viral references Video processing pipeline (trim/downsample/thumbnail + cloud object storage) Next.js dashboard for analysis, comparison views, viral exploration, and chat Gemini-powered chat layer grounded in each video’s analysis + viral reference patterns to provide strategy recommendations

Challenges we ran into

Normalizing videos from multiple sources (uploads + social links) into a consistent analysis format. Balancing technical outputs with creator-friendly explanations. Making chat responses both grounded in data and practically useful for editing decisions. Handling edge cases in async processing, model calls, and comparison workflows.

Accomplishments that we're proud of

Running a complex and resource-intensive system on limited hardware. End-to-end pipeline from raw video input to interpretable engagement insights. A working “viral intelligence chat” that connects neuroscience-style metrics with real creative decisions. Comparison mode that helps creators understand why one video likely performs better than another. A clean dashboard experience that makes complex analysis approachable.

Built With

• fastapi
• ffmpeg
• gemini
• neon
• next.js
• postgresql
• python
• react
• typescript