Momentum

Cognitive Translation Layer: verbal chaos → structured execution. Built for the USAII Global AI Hackathon 2026 | Undergraduate Track (AI for Life, Learning & Work)

The Problem: Overwhelmed Explorer Syndrome

High-potential individuals stall because abstract goals generate cognitive paralysis. Existing structured tools (like Notion) require users to already think in categories. Generic AI chat tools produce 47-step monolithic to-do lists that increase decision anxiety.

The Solution

Momentum is not a task generator; it is a Cognitive Translation Layer. It takes unstructured, emotionally overwhelmed human syntax and translates it into structured, objective milestones with a human-in-the-loop verification process.

The 3-Stage Pipeline:

1. Screen 1: Brain Dump - Raw verbal chaos input. Evaluated locally on the Edge via Transformers.js to extract sentiment and metadata without hitting the cloud.
2. Screen 2: Ambiguity Ledger - The System surfaces assumptions it must make to proceed. Hard Gate: The user must manually confirm or edit every assumption before proceeding. (Responsible AI Design).
3. Screen 3: Action Horizon - Generates broad 30/60/90-day trajectory milestones and exactly ONE 15-minute actionable micro-task.

Tech Stack & Architecture

Frontend (Client & Edge)

• Framework: Next.js (App Router) + React
• Styling: Tailwind CSS
• Edge AI Preprocessing: Transformers.js (Xenova/all-MiniLM-L6-v2) via Web Workers for zero-latency local sentiment extraction.
• Resilience: Built with zero-crash fallback architecture. If the Cloud AI layer times out (e.g., 503 errors), the UI seamlessly falls back to a locally mocked Demo Mode to ensure presentation continuity.

Backend (Cloud AI Layer)

• Framework: FastAPI (Python)
• AI Models: Gemini API (gemini-2.5-flash-lite)
• Architecture: Stateless REST API receiving Edge-processed metadata and raw text to orchestrate LLM reasoning.

How to Run Locally

1. Start the Backend (FastAPI)

cd backend
# Create and activate virtual environment
python -m venv venv
# Windows
.\venv\Scripts\activate
# macOS/Linux
source venv/bin/activate
# Install dependencies
pip install -r requirements.txt
# Run the server
python -m uvicorn main:app --reload

The backend will run on http://localhost:8000

2. Start the Frontend (Next.js)

cd frontend
# Install dependencies
npm install
# Start the development server
npm run dev

The frontend will run on http://localhost:3000

---

Responsible AI Guiding Principles

• No Prescriptive Language: The AI is system-prompted to avoid "you must" or "the correct steps are".
• Human Sovereignty: The AI never assigns a task. It proposes options, and the user must explicitly click "Approve & Claim Task" to lock it into their session.
• Transparent Uncertainty: AI confidence scores are surfaced immediately, powered by the Ambiguity Ledger highlighting exactly what data the AI is missing.

---