Case Study · Human-AI Interaction · Voice Prototype
BrainPub Project Overview
I designed and evaluated a three-agent conversational system that helps early-stage founders move from a vague idea to one concrete validation step. The HAI work was in the structure: role-separated dialogue, constrained handoffs, and a final memo artifact that turned talk into follow-through.
Disclaimer
I handled the full project end-to-end, from research and problem framing through product design.
Impact in Numbers
ESE
3.48 → 3.89
OIC
3.50 → 3.67
PBC
4.29 → 4.93
The Context
Early-stage ideas often exist before they are ready to test. A founder may have a pain point, a possible solution, or a market hunch, but those pieces are not always separated clearly enough to decide what should happen next.
The Challenge
The challenge was to design AI support for the stage before execution, where the user may not yet know what kind of help to ask for. Generic advice, startup frameworks, and open-ended AI responses can provide information, but they do not necessarily help a founder choose which uncertainty to address first.
What We Learned
The research phase showed that the core support gap was not access to information alone. The stronger need was access to structured interpretation: help making sense of a vague or partially formed idea, separating the problem from the solution, and deciding what should be tested first.
What I Designed
Three interaction design moves that made BrainPub conversational, structured, and actionable.
1) Voice-first idea exploration
BrainPub starts with onboarding and an idea submission, where the user provides initial information about their idea.
Scroll to see more →
2) Structured multi-agent system
BrainPub splits the conversation across three specialists: Sara clarifies the customer and pain, Mark narrows the direction, and Alex turns that choice into a validation plan.
Scroll to see more →
3) Validated for clarity and action readiness
The strongest reliable shifts appeared in Entrepreneurial Self-Efficacy and Perceived Behavioral Control after one guided BrainPub session.
Scroll to see more →
How I Validated It
A prototype study focused on short-term change after one structured BrainPub session.
BrainPub was evaluated through a single-group pre/post study with 16 retained participants. Each session followed the same sequence: onboarding, pre-session survey, idea submission, three guided AI phases, a generated Founder's Memo, and post-session survey. The primary measures were Entrepreneurial Self-Efficacy, Opportunity Identification Competence, and Perceived Behavioral Control.
The Results
What changed after one guided BrainPub session.
ESE
Participants left the session feeling more capable of moving an idea toward entrepreneurial action.
OIC
Opportunity thinking moved in a positive direction, but the shift stayed more tentative than the other two constructs.
PBC
The clearest action-oriented gain: the next entrepreneurial move felt more manageable after the guided conversation.
Relevant Links
Primary materials and supporting artifacts from the BrainPub thesis