AI assistant providing immediate answers to student questions within the learning interface, improving course completion by 10%.
90% of student questions in class discussion sections went unanswered, creating barriers to understanding and causing course abandonment.
Students stuck on concepts had nowhere to get quick answers, leading to frustration and disengagement when they couldn't progress through lessons.
The discussion section relied on peer-to-peer help, but with 90% of questions unanswered, it wasn't providing the support students needed to succeed.
When students hit roadblocks without support, they simply left. Unanswered questions directly correlated with lower completion rates.
We built two versions, learning from each to find the right balance between community engagement and immediate student support.
AI agent answered questions directly in the discussion section after 30 minutes of no community response, maintaining the social learning dynamic while providing backup support.
Created a separate, always-visible option within the learning interface providing instant AI responses. Students could optionally share answers as discussion posts if satisfied.
Students posted questions in the class discussion section as usual. If no community member responded within 30 minutes, the AI agent automatically replied with a contextual answer based on the lesson content.
This preserved the community-first approach but provided a safety net for unanswered questions.
A dedicated "Ask a Question" feature became permanently visible in the learning interface. Students could ask anything about the current lesson and receive instant AI-generated responses.
Both versions use the same underlying prompt system: student question + lesson content + course context. A fast model ensure responses appear within seconds.
In V2, after receiving an AI answer, students can choose to post it to the discussion section with their name, preserving community knowledge building while making immediate help the priority.
Version 2's instant support approach drove significant gains in course completion and feature adoption.
V1 tried to preserve an idealized community dynamic that didn't match actual behavior. V2 focused on what students actually needed, immediate help when stuck.
Instant answers kept students moving forward. Waiting for perfect community responses that never came was worse than fast AI assistance.
Students weren't using discussions for questions, so integrating there limited reach. Putting help directly in the learning interface met them where they already were.