Back to projects
2024
NetPostPanel
AI content-operations platform — a RAG pipeline (research → scrape → generate → vectorize), an external API other apps call, and Auto Pilot for unattended scheduled generation
An internal AI content-operations platform that turns topic research into publishable content: a full RAG pipeline (Google Custom Search + RSS → scraping → multi-provider generation → vectorization), plus a versioned external API that powers content workflows in other projects, including the DigiSpace blog pipeline.
Key Features
- External API - a versioned
POST /api/v1/generate+/translateAPI with per-client keys and rate limits exposes the whole pipeline to other apps, sync or async with callbacks - Auto Pilot - scheduled daily runs generate content from new inspiration items or custom topics, fully unattended
- Provider-agnostic AI - Gemini and OpenAI sit behind a driver abstraction, so swapping models is config, not refactoring
- Semantic pipeline - Qdrant vector search deduplicates and clusters incoming stories before they reach generation
- Autonomous aggregation - RSS auto-discovery and Google Custom Search plus scraping turn a bare topic into a self-feeding content source
- Observable LLM calls - every generation is traced in Langfuse: tokens, latency, cost
- Role-based access - admin/content manager/editor roles, with every query scoped to its owner unless admin
Tech Stack
- Backend: Laravel,
laravel/aiagents - Frontend: Inertia.js + Vue 3 (Search & Creation SPA, TipTap editor)
- AI: Multi-provider (Gemini/OpenAI) behind a driver abstraction
- Research: Google Custom Search + RSS aggregation, page scraping
- Vector DB: Qdrant for semantic search
- Observability: Langfuse for LLM tracing
- Queue: Laravel Horizon
- Tests: PHPUnit/Pest + Cypress E2E
Like what you see?
Tell me about your project — I'll propose an architecture and estimate within 24 hours.