All projects

Project 02, AI and full-stack

EduBridge AI.

Live English lectures, explained in the student's own language while the lecture is still going on.

Built to stay up during a lecture175automated tests, with automatic failover between AI providers
Timeline
Jul to Aug 2026
Event
TetraTHON 2026, Indo-French AI Innovation Sprint
Result
5th Place, 5th of 44 finalist teams
Tests
155 unit and 20 end-to-end

01, The problem

Missing one word can cost the next half minute.

Research on second-language comprehension suggests that a person needs to know about 98% of the words to follow comfortably. In a fast technical lecture, many students in English-medium colleges fall short of that.

The damage is not the one word. A student stops to decode an unfamiliar term and loses what the teacher said next, and this keeps adding up through the lecture.

Translation alone does not fix it, because a hard technical term is just as hard in Hindi. What helps is an explanation at the right level, at the moment it is needed.

02, What we built

One button, and the lecture appears in your language.

There is no account, no login and no setup form. The student presses one button, and EduBridge AI shows the lecture in English and in their own language side by side.

Three ways in

A live microphone, a shared browser tab or an uploaded recording, all handled by the same pipeline.

Two aligned columns

Verbatim English on one side and the student's language on the other, aligned paragraph by paragraph.

Tappable terms

Technical terms stay in English, as in exams and textbooks. Tapping one explains it in both languages.

An "I'm lost" button

It summarises the last 90 seconds in three points without pausing the lecture.

03, How it works

From sound to a simpler explanation, as the lecture goes on.

  1. Capture

    The browser records audio, cuts it into short pieces at natural pauses and streams them over WebSockets.

  2. Transcribe

    A speech recognition model turns each piece into English text.

  3. Find the hard words

    Technical terms are detected on the server, and the subject is inferred from the terms heard so far.

  4. Explain

    A language model rewrites the lecture in the student's language, at a level inferred from how they use the app.

04, Results

Built to keep working in a real classroom.

175

automated tests, 155 unit tests and 20 end-to-end browser tests

3 providers

Speech and language models are routed across Sarvam, Groq and Cerebras, with automatic failover if one goes down.

5th of 44

5th Place, 5th of 44 finalist teams, at TetraTHON 2026, Indo-French AI Innovation Sprint.

05, Recognition

Built during a 36-hour sprint.

  • 5th Place, 5th of 44 finalist teams. TetraTHON 2026, Indo-French AI Innovation Sprint, a 36-hour sprint at Navrachana University with ISEN Méditerranée in France, under the IndiaAI Mission.
  • Selected from the earlier round. Our selection-round project was StudySetu, an adaptive revision platform.

06, Certificates and press

The paperwork behind it.

  • Hackathon or competition

    TetraTHON 2026, 5th Place

    Navrachana University, Vadodara, with ISEN Méditerranée, France, Aug 2026

    Our team, The Caffeine Clause, won 5th Place, 5th of 44 finalist teams, at TetraTHON 2026, Indo-French AI Innovation Sprint, a 36-hour sprint at Navrachana University, Vadodara, with ISEN Méditerranée, France.

    View: TetraTHON 2026, 5th Place

Built with

  • React 19
  • TypeScript
  • Node.js
  • Fastify
  • WebSockets
  • Speech recognition
  • LLMs
  • AWS Lightsail