Vidya Base

Inspiration

The idea for Vidya Base came from a personal struggle with learning from long lecture videos. I often found myself constantly scrubbing back and forth through recordings just to capture important points, then manually writing notes to keep things organized. I wanted a faster, smarter way to turn lectures into structured notes automatically — and that became the seed for this project.

What it does

Vidya Base takes any uploaded video, extracts and chunks its audio, transcribes each segment using Gemini Nano (Prompt API) running locally in Chrome, and then organizes the summarized text into a clean, topic-based knowledge base. Everything happens directly in the browser, with no cloud involvement or external APIs.

How we built it

Vidya Base was built using React and Next.js for the frontend and logic handling. Audio extraction and chunking were done with FFmpeg, while transcription and summarization were powered through the Chrome Gemini Nano API (Prompt API). The UI loading states were handled using React components to give users a sense of progress through each stage of processing.

Challenges we ran into

The biggest challenge was handling large video files efficiently. Since the Gemini Nano model can only process around 30 seconds of audio at a time, I had to build a chunking pipeline and synchronize multiple transcription calls to ensure smooth and complete coverage of the video content. Managing performance and user feedback during these long-running processes was also tricky.

Accomplishments that we're proud of

I’m proud that Vidya Base is a fully on-device AI application — all processing happens locally, ensuring total privacy. It’s also my first app that integrates real AI capabilities directly in the browser, without relying on external APIs or servers. Achieving that level of autonomy and performance on the client side was a big milestone.

What we learned

Building Vidya Base taught me how to integrate on-device AI models effectively using the Chrome Gemini Nano API, as well as how to manage browser performance when working with large media files. I learned a lot about chunked processing, asynchronous pipelines, and creating clear, user-friendly progress feedback. Most importantly, I discovered how capable client-side AI has become — making it possible to deliver privacy-first intelligent tools entirely in the browser.

What's next for Vidya Base

Next, I plan to add the ability for users to pause transcription midway and save their progress in local storage, making long video processing more flexible. I also want to expand support to audio files and live recordings. Vidya Base is open-source, and I’m excited to see others build upon it to create even better on-device AI experiences.

Built With

• nextjs
• react