Actual participant evidence
Direct excerpts, findings, or observations from real research—with source references.
Future direction
How this could get better
This demonstration uses publicly available information to construct synthetic, inferred perspectives. A production version could become substantially more useful by connecting the same framework to an organization’s real research repository.
Research repository
- Study reports
- Interview transcripts
- Research videos and timestamped clips
- Coded findings and themes
- Survey findings
- Usability sessions
- Customer quotes
- Historical insights
01Evidence retrieval
Find related material for the question.
02Archetypes + synthetic participants
Keep declared perspective and generated participant distinct.
03Question + LLM
Generate a bounded synthetic response.
ResponseEvidence, sources, and confidence included
Actual research as grounding
Connect answers to what people really said.
The retrieval layer could bring forward relevant studies, transcripts, clips, quotes, surveys, reports, and coded themes without treating every source as equally strong.
Public-data demonstration
“This power user would probably want more control.”
Research-grounded direction
“This is consistent with three prior studies in which experienced builders described control and visibility as important. Here are the underlying excerpts.”
Keep the boundaries visible
One answer, four different kinds of knowledge.
Archetype-level inference
Declared behavioral characteristics used to model a point of view.
Synthetic response
A generated reaction to the current question, clearly labeled as hypothetical.
Unknown / unsupported
Gaps the available material cannot answer and should not be smoothed over.
Long-term direction
From hypothesis demo to research intelligence.
A future version might first answer, “What do we already know about this question?” and then recommend, “What should we research next?” That would turn the framework into a bridge between an organization’s accumulated evidence and its next learning decision.