01
The problem
Product managers and founders spend weeks researching competitors, mining customer pain points, sizing markets and synthesising findings before making go or no-go decisions. The process is slow, biased and often driven by instinct rather than structured analysis.
02
The approach
Built a multi-agent system with specialised roles for market landscape analysis, customer pain research, opportunity sizing, risk assessment, strategy synthesis and quality audit. Each agent uses an explicit framework and produces structured outputs for the final report.
03
Challenges
Generic prompts produced shallow analysis, so expert personas and explicit constraints were added. A quality-audit agent flags unsourced claims. The pipeline was also changed to collect every agent's output rather than only the final task.
04
What I learned
Expert framing produces stronger output than generic instructions. Explicit reasoning structures improve reliability. Triangulating across sources catches weak claims early, and a final quality gate is essential.
05
Outcomes
Reduced a multi-week discovery process to minutes. The system automates competitor research, customer-pain mining and market sizing, producing structured reports with risk analysis and practical roadmaps.
Tech stack
- CrewAI
- Python 3.12
- OpenAI GPT-4o
- Serper API
- Reddit API
- Gradio