github.com/Nitrajsinh-Solanki/Meeting-Genius
06 / 11
🤖 Hackathon Submission
Meeting Insights Generator
Built for the 'From Words to Worlds — Multimodal GenAI' hackathon, this app handles the complete meeting intelligence pipeline. Upload a recording and AssemblyAI transcribes it to full text. The Facebook Bart Large CNN model on Hugging Face then summarizes the transcript. Key points from the summary drive Pexels image searches for visual highlights. Finally Deepgram converts the summary to an audio file, and everything is stored persistently on Cloudinary.
Engineering Case Study
- Challenge
- Reviewing a long meeting recording manually takes time, and a plain transcript does not provide a concise, listenable summary with supporting visuals.
- Approach
- Connect AssemblyAI transcription, Hugging Face’s BART summarization model, Pexels image search, Deepgram text-to-speech and Cloudinary storage in one React/Vite and Node/Express application.
- Contribution
- Integrated the audio upload-to-output pipeline: transcription, text summarization, visual highlights matched to key points, spoken summaries and persistent storage.
- Outcome
- Delivered a multimodal workflow that turns meeting audio into text, a concise summary, related imagery and audio playback for the hackathon submission.
Key Features
- Audio upload → full text transcription via AssemblyAI
- AI summarization using Facebook Bart Large CNN (HuggingFace)
- Visual highlights — Pexels images matched to key summary points
- Text-to-speech audio output of summary via Deepgram
- All generated content stored persistently on Cloudinary
- Full multimodal pipeline: speech → text → summary → image → audio
- React + Vite frontend with Node.js/Express backend
Tech Stack
ReactViteNode.jsExpress.jsMongoDBAssemblyAIHuggingFace Bart CNNPexels APIDeepgram APICloudinaryJWT
Links
Award
Prize
Submission
Event
From Words to Worlds — Multimodal GenAI