Runtime terror

Inspiration

Meetings are often long and information-dense, making it hard to track key takeaways or decisions. We wanted to create a privacy-first AI assistant that could transcribe, summarize, and help users interact with meeting content in real time—without relying on cloud services or compromising sensitive data.

What it does

Nexora is an AI-powered meeting companion that:

Transcribes meeting audio to text using Whisper
Summarizes key points with a local LLM (LLaMA 3)
Extracts an actionable checklist with owners, due dates, and status
Lets users ask questions about the meeting via an interactive chat interface
Works offline or locally using Ollama and Flask

How we built it

Frontend: Built with React + Tailwind CSS + Framer Motion for a modern, responsive UI.
Backend: Flask APIs handle file uploads, transcription, summarization, and chat logic.
AI Models: Integrated Whisper for transcription and LLaMA 3 via Ollama for summarization and contextual chat.
Storage: Used Google Cloud Storage for optional transcript uploads, with local fallback for development.
UI/UX: Includes animated chat bubbles, Apple-style loading spinners, download options, and seamless navigation.

Challenges we ran into

Integrating local LLMs (Ollama) with the frontend chat experience in real-time.
Formatting the AI responses consistently into summaries and structured checklists.
Managing shared state across transcript, summary, and chat sections with clean UX.
Ensuring everything works smoothly offline while keeping the UI responsive.

Accomplishments that we're proud of

Seamless AI pipeline from audio to transcript to summary to chat—all on local hardware.
Elegant UI with rich feedback (typing animations, loaders, downloadable text).
Modular architecture—each feature is its own component and can be reused across other apps.
Completed a fully working MVP within hackathon constraints with near-perfect transcript and summary accuracy.

What we learned

How to integrate Whisper and LLaMA 3 locally with real-time web apps.
Frontend/backend communication for model interaction using Flask and REST.
Creating privacy-first AI solutions that don’t rely on OpenAI or external APIs.
Designing with user-first principles—clear actions, feedback, and smooth transitions.