CPAI Platform Documentation
CPAI (Cognitive PDF AI) is a self-hosted, CPU-based AI document intelligence platform built for the AI Essentials demo. It enables users to upload PDF documents, automatically extract and vectorize content using a RAG (Retrieval-Augmented Generation) pipeline, and generate comprehensive AI-powered summaries — all running on commodity hardware without a GPU.
What CPAI Does
- Upload a PDF document through the web interface
- Extract text content from the PDF
- Chunk the text into overlapping segments and generate vector embeddings
- Store chunks with embeddings in PostgreSQL (pgvector)
- Summarize using RAG — retrieve relevant chunks, assemble a prompt, and generate a structured summary via Qwen2.5 LLM
- Chat — ask follow-up questions about the document using the same RAG pipeline
Key Features
| Feature | Description |
|---|---|
| PDF Upload | Drag-and-drop, up to 50 MB |
| RAG Pipeline | Automated text extraction, chunking, embedding, and summarization |
| Interactive Chat | Ask questions about uploaded documents |
| JWT Authentication | Secure login with role-based access |
| Audit Logging | All operations tracked with user, action, timestamp, IP |
| 7 Languages | CS, EN, DE, FR, ES, RU, ZH |
| Real-time Metrics | Service health, response times, success rates |
| PDF Export | Export summaries as branded PDF reports |
| Dark/Light Theme | Toggle between dark and light modes |
Architecture at a Glance
Quick Links
- Live Demo: cpai.microsaasplatform.com
- API Swagger: api.cpai.microsaasplatform.com/swagger
- Default Login:
hpe/hpe
Technology Stack
| Layer | Technology |
|---|---|
| Frontend | React 19, Vite 7, TailwindCSS 4, TypeScript |
| Backend | .NET 10, ASP.NET Core, EF Core |
| Database | PostgreSQL 16 + pgvector 0.7 |
| Workflows | n8n Community Edition |
| LLM | Ollama + Qwen2.5-1.5B (CPU) |
| Deployment | Coolify on Hetzner (64 GB RAM, no GPU) |