Infochat

Context

Infochat is Dreamension's second product. Each streamer gets their own Chrome extension, customized to their colors. It notifies viewers of new content, brings their social networks together and displays information about the game being played. A chatbot answers the recurring questions of the chat. The data is hosted on servers in France.

What I did

The heart of the project is a script that generates the extensions automatically. It takes as input the streamer's Twitch identifier, their design preferences and the features chosen. It then produces the complete code, manifest, background scripts and interface. The extension arrives preconfigured, which simplifies onboarding.

I also designed the analysis of the questions asked in Twitch chats. It identifies recurring themes and feeds the chatbot's FAQ. The processing relies on approximate string matching, completed by Python analyses, word clouds, sentiment analysis and keyword extraction.

The collection of streaming data through the Twitch API finally produced statistics for the streamers, games played, audiences and peak hours.

Outcome

Automated generation removed the manual development of each extension. The chatbot was still imperfect, approximate matching was showing its limits with a volume of data that remained small. That observation shaped what came next, collect more before going further.