Preguntas frecuentes
Goal
Built a full-stack AI media platform that generates, verifies, and publishes wellness content at scale — designed to operate like a premium publication (NYT Health, WSJ) but at 1/250th the cost. 86 API endpoints, 11 specialized agents, 175 UI components, and a proprietary Evidence-Centered Content (ECC) system that scores every article on a 1000-point scale before publishing.
Why
Traditional digital publishing is broken: hiring writers costs $50K+/month for volume, fact-checking is manual and inconsistent, and health content carries regulatory risk (FTC/FDA). LIME solves this by treating content as a pipeline — AI generates, agents verify, algorithms score, humans approve. The vision: a media company that scales like software.
What worked
Master Publishing Algorithm (MPA):** 10-variable weighted scoring (0-1000) covering topic value, content quality, SEO readiness, fact-checking, legal compliance, brand voice, visual assets, timing, distribution, and monetization
Health Claims Verification:** Automated FDA/FTC compliance detection with 5 evidence tiers (RCT, Meta-Analysis, Observational, Preclinical, Anecdotal) and character-position mapping for inline claim tracking
Signal Intelligence Pipeline:** 34-source monitoring with Topic Value Score (TVS) algorithm, signal deduplication via content hashing, and urgency classification (Immediate → Evergreen)
Thick Text Engine:** 5-dimension reader engagement scoring (Prediction, Agency, Rhythm, Repair, Continuity) with AI-powered insert suggestions and A/B testing framework
Agent Orchestration:** 11 specialized agents (Research, Writer, Editor, SEO, Social, Legal, Packaging, Scout, Scoring, Publisher, Synthetic Critic) with human-in-the-loop checkpoints and full audit trails
Production-ready:** 35K+ lines TypeScript, Prisma schema with 20+ models, role-based access (Admin → Writer), deployed on Vercel
Challenges
Backend infrastructure at 100% but UI integration incomplete — dashboards exist but wiring to live data is partial
Agent ecosystem designed but autonomous execution not fully tested in production workflows
Health claims detection needs expansion beyond current 25 claim categories
Signal sources defined (34) but real-time ingestion pipelines need API integrations
Thick Text scoring implemented but A/B testing framework lacks sufficient production data
Next steps
Wire TransparencyDashboard and PipelineTracker to live workflow data
Complete agent execution loop with cost/token monitoring per workflow
Expand health claims database with PubMed/DOI citation verification
Build real-time signal ingestion for top 10 sources (FDA, PubMed, industry news)
Deploy Thick Text A/B testing with engagement metrics collection
Add GraphRAG knowledge graph for cross-article entity relationships
Tools
Tech:** Next.js 14, TypeScript, Prisma, SQLite/PostgreSQL, Tailwind CSS, Claude API (Sonnet 4), NextAuth.js, Compromise NLP, Vercel
Scale:** 86 API routes | 175 components | 11 agents | 35K+ lines TS | 8 content types | 11 categories | 34 signal sources Economics:** ~$0.50-2.00/article vs $100-500 human-written | ~$200/month operational vs $50K+ traditional
