Reference prices—not fixed quotes
Within Leaf Technology’s current engagement scope, AI / RAG systems commonly reference NT$100,000–300,000+. Clean documents, simple users, and no complex permissions can approach the low end; connecting databases, multi-department permissions, admin backends, and existing login pushes cost up.
An AI PoC can start from NT$50,000 to validate “can these documents be answered”—not to promise a full product.
What drives cost up?
Poorly scanned PDFs, OCR for scans, contradictory policies, requiring cited and reviewed answers, and writing back to ERP or support systems. Those are scope—not the model unit price alone.
What does reasonable acceptance look like?
Prepare 20–50 real questions with expected sources. Accept on whether the right document was found, whether the answer distorts the original, and whether unauthorized users can’t ask. Don’t use “99% accuracy” with no question bank as a contract clause.
Author
葉科技
With 13+ years of software development experience, we focus on AI system integration and Web systems. Founded in 2022, our work includes our own AI video SaaS (Mofly AI), a personality-quiz Web product (MindTest AI), an enterprise AI assistant (Leaf AI Workspace), membership and inventory systems and admin backends, plus embedding recognition capabilities into existing products.
FAQ
Does the fee include model monthly costs?
Usually not. API, vector store, and cloud hosting are operating costs, estimated separately in the proposal.
Next, see RAG Development Services.