CortexFlow

Loading / CortexFlow screen
Voice and text input interface for capturing samples and initiating real-time cognitive analysis.
Cognitive dashboard showing domain-level analysis, biomarker trends, and real-time cognitive load insights.
Interactive 3D brain workspace mapping linguistic signals to region-level cognitive activity and indicators.

Inspiration

MnM is a two-person team built by Manika Kutiyal and Aditya Verma. We wanted to create something that goes beyond a typical AI demo and solves a real problem in a clear, responsible way.

CortexFlow was inspired by the idea that meaningful cognitive signals can already exist in how people speak and write. We wanted to build a system that can turn that raw input into something structured, interpretable, and useful for fast screening workflows.

What it does

CortexFlow is a full-stack cognitive signal analysis platform that accepts typed text or recorded speech and turns it into a structured report.

It:

Captures browser-based voice input or typed text
Uses Gemini-compatible APIs for transcription
Extracts deterministic linguistic signals across lexical, semantic, prosodic, syntactic, and affective dimensions
Maps those signals into interpretable brain-region level outputs
Streams a safety-framed, non-diagnostic report
Visualizes biomarker intensity in an interactive 3D brain workspace
Supports authentication and report history with Firebase and Supabase

How I built it

We built CortexFlow as a monorepo with a clear split between frontend and backend.

The frontend is built with Next.js, React, TypeScript, Tailwind CSS v4, Recharts, and Three.js. It handles the UI, auth flow, audio capture, report history, and proxy API routes.

The backend is built with FastAPI and Python. It receives normalized input, computes deterministic biomarkers, derives cognitive load signals, and returns the final structured report.

For AI integration, we used Gemini-compatible APIs for transcription and report-safe summary generation. Firebase and Supabase handle user authentication and session persistence. The deployment setup was designed to be hackathon-friendly, with a Vercel frontend and a container-ready backend.

Challenges I ran into

One of the biggest challenges was keeping the system both technically impressive and responsibly framed. Since the project works with cognitive analysis, we had to make sure the output stayed non-diagnostic and safe.

Another challenge was balancing AI-generated processing with deterministic logic. We wanted the final report to be interpretable and reliable, not just a black-box output.

We also had to make the whole repo easy to run for judges, which meant keeping the architecture clean, supporting local demo mode, and making sure the auth bypass and backend wake flow worked smoothly.

Accomplishments that I'm proud of

We are proud that CortexFlow is not just a prototype, but a complete working system.

Some of the things we are most proud of are:

Building a full-stack pipeline from speech and text to structured cognitive analysis
Creating a clear frontend-backend architecture that is easy to understand and extend
Designing a 3D brain visualization that makes the results more intuitive
Keeping the output safety-framed and non-diagnostic
Making the project hackathon-ready with a clean demo path and deployment-friendly setup

What I learned

This project taught us how to connect AI, backend logic, and interactive frontend design into one coherent product.

We learned how important it is to:

Keep AI outputs structured and explainable
Design around real user experience, not just technical novelty
Build systems that are both demo-friendly and scalable
Think carefully about responsible framing when working with sensitive problem spaces

It also helped us improve our skills in full-stack development, API integration, visualization, system design, and fast iteration under hackathon constraints.

What's next for CortexFlow

Next, we want to make CortexFlow even more useful, accurate, and adaptable.

Some possible directions are:

Expanding support for more speech and text analysis workflows
Improving the biomarker mapping and report explainability
Adding more domain-specific screening templates
Enhancing the 3D brain visualization and interaction model
Building stronger export, collaboration, and longitudinal history features
Exploring integrations with educational, wellness, and accessibility-focused applications

Our long-term goal is to keep evolving CortexFlow into a practical cognitive signal platform that is easy to use, easy to understand, and meaningful in real-world workflows.