Artificial intelligence · EdTech

TeachUNITED's knowledge, answered by AI in any language

AI assistant (RAG) for TeachUNITED's teaching staff: it indexes documents, audio, and video, and answers questions from the web and mobile apps, in any language.

TeachUNITED · TUbot

TeachUNITED is a U.S. nonprofit that brings world-class teacher training to schools and partner organizations across different countries. Its challenge: all that knowledge —guides, lessons, and training materials in PDF, Word, and PowerPoint, plus videos and audio— was scattered, making it hard for each teacher to find the right answer at the right moment.

We built TUbot: a web platform with a backend where TeachUNITED staff upload their material to a RAG (Retrieval-Augmented Generation) system. The system indexes everything in vector databases and automatically transcribes videos and audio to make them searchable and multilingual. Content stops being loose files and becomes a living knowledge base.

The result is an AI-powered chat —available on the web and on iOS and Android mobile apps— that answers teaching staff's questions with TeachUNITED's real information, citing the source resources and in each user's language. It is currently deployed in Mexico and Peru, and soon in Senegal.

1

Upload

Staff upload PDF, Word, PowerPoint, audio, and video.

2

Index

The system transcribes and stores everything in vector databases (RAG).

3

Ask

Teachers ask via chat from the web or the mobile app.

4

Answer

The AI replies in their language, citing the source resource.

Upload PDF, Word, PowerPoint, audio, and video to the knowledge base

Indexing with vector databases (RAG)

Automatic transcription of videos and audio

Multilingual answers in each user's language

AI chat that cites the source resources

Available on web and mobile apps (iOS and Android)

Admin panel for the TeachUNITED team

Upload a PDF or a video, and the AI already knows the answer

Staff upload documents, presentations, audio, and video; TUbot transcribes them, indexes them in vector databases, and makes them multilingual. The AI answers by citing the real source, without making things up.

If you can imagine it, we can build it.

Tell us what you need. We'll get back with a clear proposal, no jargon.