Top Dog

An AI agent that monitors competitors, scores signal relevance, and autonomously rewrites its own prompts when scores drop—true self-improving competitive intelligence.

Comment

Inspiration

We were inspired by the challenge of building truly autonomous AI agents that learn and improve. Competitive intelligence is a real pain point—companies spend hours manually tracking competitor changes and often miss critical signals. We wanted to build an agent that autonomously monitors competitors and gets smarter over time without human intervention.

What it does

TopDog-Competitor-WebWatcher is an autonomous competitive intelligence agent that monitors competitor websites, detects strategic changes (pricing, features, messaging), and teaches itself to be better by rewriting its own detection prompts when it produces low-relevance signals.

How we built it

Built in ~2 hours using Node.js/TypeScript backend and React frontend: Backend: Agent core with Gemini integration for URL analysis, Braintrust for scoring and self-improvement, Lightdash for visualization, Express API Frontend: React dashboard with real-time polling, Recharts for visualization, dark theme UI Architecture: User triggers agent → Gemini analyzes URLs → Braintrust scores → Self-improvement triggers if needed → Lightdash visualizes → Dashboard updates Tech stack: Google DeepMind (Gemini), Braintrust, Lightdash, Node.js, TypeScript, React, Vite

Challenges we ran into

API quota limits — solved with mock data fallback for demo reliability Model name issues — discovered correct Gemini model names through API queries Real-time updates — implemented dynamic polling (1s when running, 5s normally) Self-improvement visibility — added version tracking and visual indicators Time constraints — focused on MVP and core differentiator Accomplishments that we're proud of

Accomplishments that we're proud of

True autonomous self-improvement — agent rewrites its own prompt file 3 sponsor tools integrated — Google DeepMind, Braintrust, Lightdash Real-time dashboard — visually impressive, live updates Graceful fallbacks — works even when APIs are rate-limited Complete MVP in 2 hours — fully functional, demo-ready Clean architecture — maintainable and extensible Visual polish — modern, professional dashboard

What we learned

Self-improvement requires careful design but is powerful API quotas need fallbacks for reliable demos Real-time updates improve engagement Each sponsor tool has unique strengths to leverage MVP focus beats feature bloat in time-constrained environments Mock data fallbacks ensure demos always work

What's next for Top Dog

Production API integration — real competitor monitoring More competitors — dynamic URL addition Alert system — email/Slack notifications Historical analysis — trend analysis over time Multi-user support — team collaboration Advanced self-improvement — Braintrust Loop integration Database migration — PostgreSQL for scalability Scheduled monitoring — configurable schedules Export features — CSV/PDF reports Mobile app — React Native for on-the-go monitoring Vision: Transform TopDog into a comprehensive competitive intelligence platform powered by autonomous, self-improving AI agents.

Built With

  • braintrust
  • cursor
  • deepmind
  • google
  • lightdash
Share this project:

Updates