Inspiration
The inspiration for Katha Khoj was mainly due to our own realization that we couldn't answer our friends' questions about Sikhi in an effective manner. The way we found love for the Guru was through listening to Katha, and so we wanted to make Katha from the most renowned Sikhs of the 20th century and of today easily accessible.
What it does
Katha Khoj is a semantic search engine that provides answers not through AI-generated responses, but by utilizing AI to sort through thousands of hours of Katha and providing timestamps to videos and easy access to answers through the click of a few buttons.
How we built it
We utilized our knowledge in AI to run scripts in order to collect data. Once that data was collected, we used agentic RAG to understand queries, retrieve the relevant recordings, and provide direct citations through built-in video and audio players. The scripts we wrote first transcribed the audio in Punjabi using Eleven Labs. We then translated those Punjabi transcripts into English using the Gemini API.
Challenges we ran into
We ran into several significant technical and design challenges, including:
- Developing strategies to optimize and lower API costs.
- Addressing search latency to ensure queries were efficient and not too time-consuming.
- Creating an intuitive, modern UI design that remains user-friendly.
Accomplishments that we're proud of
We're proud of the progress we were able to make in the span of just one weekend. We far surpassed our initial MVP feature list and were able to implement advanced features that we originally thought were much farther down the road.
What we learned
We learned how to maximize our manpower in the most effective way possible thanks to AI tools like Claude Code and several others. This allowed us to build lasting solutions and a clear roadmap for future success.
What's next for Katha Khoj
Katha Khoj aims to be the future for educating people wherever they may be on their Sikhi journey. We hope to become sustainable in the long term, which will require further improvements and performance optimizations.
Built With
- azure
- azure-ai-search-and-azure-blob-storage-for-cloud-services
- claude-(anthropic)-for-ai-powered-answer-synthesis
- django
- django/python-on-the-backend
- elevenlabs
- elevenlabs-for-punjabi-speech-to-text-transcription
- langgraph
- postgresql
- postgresql-and-redis-for-data-storage
- python
- react
- redis
- typescript
- vite
Log in or sign up for Devpost to join the conversation.