Weave

Overview

Weave is a production-grade document processing platform powered by Adobe PDF Embed API. The project features a comprehensive backend with multiple AI services distributed across multiple services and a modern React/Next.js frontend.One of the core features of the Weave platform is its ability to perform relevance extraction without requiring an internet connection. This is achieved through a robust semantic search mechanism that operates entirely offline. This project was built for final submission of Adobe Hackathon 2025.

Demo Videos

For a quick overview of the Weave platform, check out our demo videos:

• Short Demo (2 mins)
• Complete Walkthrough (10 mins)

Documentation

• Changelog: How we used 1a and 1b to build the final backend?
• Production Features: Production features and capabilities
• Observations: Performance insights and recommendations
• Podcast Service: Podcast Generator API, its features, and usage instructions.
• Chunk Builder: Build Chunks + Geometry + Embeddings
• Embeddings: Embedding process using sentence-transformers.
• Outline Extractor: Heading discovery and sanity checks.
• Persona Service: Persona-based content filtering and personalization.
• Relevance Service: Document connection and relationship analysis.
• Chat Service: Chatbot API, its features, and usage instructions.
• Insights Service: Insights Generator API, its features, and usage instructions.
• Summary Service: Collection Summaries Generator API, its features, and usage instructions.

Build Commands

# Build the Docker image
docker build --platform linux/amd64 -t hackathon-app .

Docker Deployment

Run the application using our preferred Docker run command:

# Run the application
docker run --rm --name hackathon-app \
  -p 8080:8080  \
  -p 8000:8000 \
  -e ADOBE_EMBED_API_KEY=YOUR_ADOBE_EMBED_API_KEY \
  -e GEMINI_API_KEY=YOUR_GEMINI_API_KEY \
  -e LLM_PROVIDER=gemini \
  -e GEMINI_MODEL=gemini-2.5-flash \
  -e TTS_PROVIDER=azure \
  -e AZURE_TTS_KEY=YOUR_AZURE_TTS_KEY \
  -e AZURE_TTS_ENDPOINT=YOUR_AZURE_TTS_ENDPOINT \
  hackathon-app

⚠️ Important Note: Exposing port 8000 is crucial for the conversational (Two AI Speakers) podcast generation feature to work properly. Without port 8000, the Azure TTS chunks will fail to reach the frontend due to CORS issues and proxy problems. While the podcast will still be generated, it will remain inside the Docker container at podcast_service/audio_data/ and won't be accessible from the frontend. To experience our flagship podcast feature, make sure to include -p 8000:8000 in your Docker run command.

Credentials

IMPORTANT: We have provided the all the credentials which was used to build the application in the following document click here, please use them to run the application in case of any issues. We will be revoking them soon after the hackathon.

Offline Capabilities

One of the core features of the Weave platform is its ability to perform relevance extraction without requiring an internet connection. This is achieved through a robust semantic search mechanism that operates entirely offline. By leveraging pre-trained models and efficient indexing, the platform ensures that document processing and analysis can be conducted securely and independently of external network dependencies.

👥 Team

Team Valhalla

• Umesh Meena - @umez-57
• Shreyas S - @zhreyu

Dev Notes

• Integration Guide: Dev log notes on backend-frontend integration
• API Error Codes: Dev log notes on complete error handling guide
• Logging System: Dev log notes on comprehensive logging documentation
• Frontend Error Handling: Dev log notes on frontend error management

Acknowledgments

Team Valhalla is grateful to the amazing people at Adobe and Unstop for providing us with the opportunity to build this project. We had great fun working on this project and learning about the Adobe PDF Embed API and the AI ecosystem.

---

For technical support or questions about the implementation, refer to the detailed documentation in the docs folder or check the specific service documentation in the backend and frontend directories.