Competitive Intelligence Agent
An AI agent that runs weeks of market research in minutes.
- Context
- Competitive research is the kind of work that is always urgent, always expensive, and always already out of date by the time it lands. I wanted a repeatable system a team could run in under 15 minutes.
- Timeline
- Delivered 2025, internal tool
- Role
- Agent design, prompt engineering, Notion integration, docs
- Research
- 5–15 min / 20 competitors
- Output
- Auto-synced Notion DB
- Reliability
- Async + retry
What I built
- A Python agent that accepts a list of competitor names and a research brief.
- Structured multi-step prompting against Google Vertex AI with grounded web search.
- Notion integration that writes 50+ structured fields per competitor directly into a team database.
- Async execution so 20 competitors complete in parallel.
- Retry and backoff logic so partial network failures do not lose a run.
- A README, requirements.txt, and config template for anyone else to run it.
Architecture
Interesting decisions
Structured output over free-form
The agent forces structured JSON output per competitor. That is what lets the Notion sync stay honest and queryable. Free-form text would have been easier to prompt and useless to consume.
Notion as the UI
Teams already live in Notion. Building a custom UI would have slowed adoption. Writing straight into their existing database meant the tool paid off on day one.
Result
What was a multi-day research exercise is now a 5–15 minute run. Output is more consistent than the human version and can be reproduced whenever the landscape shifts.
Jumbo / InterStar
Full e-commerce and ERP system. Storefront returning soon after a compliance update.
Lab Data Pipeline
Manual lab operations replaced with a self-updating cloud pipeline.
PickAxe
Data-heavy client site with a custom interactive profit calculator.