Announcement: Enabling AI Chatbot (V1) with NLWeb on Sunbird ED

[chethann007]

We are excited to announce that we are working on enabling an AI-powered chatbot within Sunbird ED to make content discovery easier and more interactive

🔹 What’s coming in V1

• The chatbot is powered by NLWeb (open-source natural language layer).
• Learners will be able to ask questions in natural language and get relevant responses from Sunbird content.
• Personalized responses based on class, medium, and subject preferences.

🔹 Target Rollout

This will be part of release-8.0.0, planned for end of September.

🔹 Future Enhancements

• Will support PDF content as the knowledge source.
• Integration with more content formats (ePub, videos with transcripts, etc.).
• Direct links to Sunbird Player inside chat responses.
• More … 🚀

🔹 Why this matters

Currently, content discovery in Sunbird is filter-driven and manual.
With this chatbot, learners and teachers can:

• Quickly find the right content by asking questions.
• Get contextual answers instead of browsing multiple filters.
• Seamlessly jump to Sunbird Player to view the full resource.

🔹 Feedback & Collaboration

This is an early-stage V1 rollout. We welcome suggestions, ideas, and contributions from the community to shape how this chatbot evolves.

📢 Stay tuned for updates in this thread as we continue improving the chatbot and expanding content support!

[pallakartheekreddy]

As part of the ongoing discussion on enabling AI/NLQ search in the Sunbird portal, I’ve drafted a detailed Design Wiki for community review.

Design Document: AI-Enabled Search for Sunbird Portal

Summary of the proposal:

• Move from strict keyword-based search (Elasticsearch) to AI-enabled semantic search.
• Leverage LLM + RAG to understand intent and retrieve contextual results.
• Convert content metadata from JSON → JSON-LD for semantic enrichment.
• Store embeddings in a Vector DB (Qdrant) and enable query handling through NLWeb.
• Maintain a seamless UI experience, aligned with the existing search flow.

Tech Stack (Proposed):

• Vector Database: Qdrant (for semantic embeddings store).
• NLWeb: For embeddings generation and storage (leveraging OpenAI).
• Flink Job: To generate JSON-LD from existing JSON metadata.

Key Benefits:

• Better content discoverability (synonyms, related concepts, context awareness).
• Scalable and future-proof metadata model (JSON-LD, schema.org).
• Extensible to support more content types like PDFs, videos, etc.

Would love for the community to:

• Review the design doc.
• Share feedback, improvements, or concerns.
• Suggest any additional considerations for V1 rollout.

Thanks 🙏