A web application that autonomously discovers innovations by finding structural bridges between unrelated fields, validates them against real developer pain, and builds working code.
- Frontend: React 19 + Vite, Firebase SDK (modular)
- Backend: Firebase Cloud Functions v2, Firestore, Storage
- Design: Dark cinematic aesthetic (#0D1117 + #4ECDC4 teal)
meridian/
src/ # React frontend
components/
shared/ # Logo, Button, CopyButton, ShareableLink
landing/ # Hero, SocialProof, ProblemComparison, HowItWorks, Footer
app/ # Header, ConversationPanel, AgentStatusFeed, OverseerBar,
# DiscoveryBoard, HistoryPanel
cards/ # ProblemCard, FrontierCard, BridgeCard, ValidationCard,
# CodeCard, MarketGapCard
launchpack/ # LaunchPack, HNPostTab, DevOutreachTab, ProductHuntTab
hooks/ # useAuth, useDiscovery, useDiscoveryHistory
pages/ # LandingPage, AppPage, DiscoveryPage
config/ # firebase.js
utils/ # agentNames.js, formatters.js
functions/ # Firebase Cloud Functions
agents/ # 10 agent modules
apis/ # 8 external API wrappers
pipeline/ # orchestrator.js
overseer/ # overseer.js
lib/ # gemini.js, groq.js, goalAnchor.js, pdfGenerator.js
# Install dependencies
npm install
# Set up environment
cp .env.example .env
# Fill in your Firebase config values
# Start development server
npm run dev- Create a Firebase project at console.firebase.google.com
- Enable Firestore, Authentication (Google provider), and Storage
- Upgrade to Blaze plan (required for Cloud Functions, still free within limits)
- Set secrets:
firebase functions:secrets:set GEMINI_API_KEY firebase functions:secrets:set GROQ_API_KEY firebase functions:secrets:set TAVILY_API_KEY firebase functions:secrets:set GITHUB_TOKEN
- Deploy:
npm run build firebase deploy
- Prompt Sharpener — Converts raw input into structured research brief
- Frustration Scanner — Searches SO/GitHub/HN for real developer pain
- Frontier Detector — Finds capabilities from the last 90 days
- Adjacent Domain Finder — Searches 10 unrelated fields for structural analogies
- Adversarial Critic — Attacks bridges with 3 rejection tests
- Reality Checker — Finds real developers who need this
- Code Translator — Generates working Python code
- Code Verifier — Independent syntax/logic/test verification
- Market Gap Analyst — Identifies competitor gaps
- Publisher — Creates GitHub repo + PDF + launch pack
A continuous intelligence that scores every agent output on:
- Trajectory (1-10): Is the pipeline heading toward useful innovation?
- Credibility (1-10): Is output backed by real data?
- Novelty (1-10): Is the innovation genuinely new?
Can STOP the pipeline if quality drops.
Built for the Agentathon 2026, April