AI PROTOTYPE · 99P Labs
Prototyping a multi-agent system for trend intelligence
Co-developed a working prototype that uses specialized agents to discover, challenge, and synthesize emerging market signals.
- Role
- Graduate innovation project, 99P Labs
- Period
- 2025
- My ownership
- Co-developed the research workflow, agent responsibilities, evidence model, and prototype documentation.
THE CHALLENGE
Trend research is often a pile of links followed by a confident summary. The harder problem is distinguishing a durable market signal from a temporary spike in attention.
THE INSIGHT
Separating discovery, validation, contradiction, and synthesis creates productive tension. The system becomes more useful when each conclusion can be traced back to evidence and challenged.
THE APPROACH
From evidence to execution.
- 01
Decompose the judgment
Assigned distinct roles for signal discovery, source validation, counterargument, and synthesis.
- 02
Preserve provenance
Designed outputs so claims retained links to supporting evidence instead of disappearing into a summary.
- 03
Prototype the loop
Built the workflow in OpenCode and documented where human judgment should remain in the system.
AGENT WORKFLOW
STRATEGIC ARTIFACT / RECONSTRUCTED
A research room, not a single oracle
Scout
Finds weak signals across product, customer, company, and cultural sources.
Skeptic
Looks for contradictory evidence, recycled narratives, and unsupported momentum.
Strategist
Connects validated signals to customer behavior and commercial implications.
THE OUTCOME
The prototype demonstrated how multi-agent design can make trend identification more structured, inspectable, and useful for product and marketing decisions.
Measurement note: This is a working prototype and learning project. It does not claim a commercial outcome.
WHAT I CARRY FORWARD
Separating discovery from challenge made the output easier to inspect because the system had to show both the supporting evidence and the objection.