AI Requirement Analysis
Want to adopt AI but unsure where to start? We first clarify workflows, data, and acceptance criteria, then decide on API, RAG, Agent, or a general LLM—and deliver a buildable scope, cost range, and PoC timeline.
Review current workflows, data sources, existing systems, and desired outcomes. Without these, you can’t tell whether you need chat, a knowledge base, or automation.
Decide on a general LLM API, RAG knowledge base, controlled Agent, or structured output only. Most enterprises do not need to build their own model.
Document PoC scope, production scope, timeline, and cost range—and list what we will not promise (e.g. zero hallucination).
After confirmation we can take a PoC or full project; you can also start with a 60-minute paid consult for a written conclusion.
時程：Free requirement assessments usually get a directional reply within a few business days; a paid 60-minute consult can produce a written path faster. PoCs commonly take 2–6 weeks; production systems are scoped separately.
費用：Free requirement assessment has no fee. For full architecture and cost judgment, choose 60-minute AI adoption consulting (NT$2,000–3,000, creditable toward later development). Production work is quoted by scope.
正式價格於需求確認後提供，以上為參考。
Leaf AI Workspace（leafflow-ai.vercel.app）：企業文件 RAG 問答、白名單工具查專案／帳款，以及通知前人工確認的公開示範。
葉科技自有 AI 影音 SaaS：劇情生成 → AI 角色 → AI 分鏡 → 圖片生成 → 影片生成 → TTS → 字幕 → 成片，做成可註冊、可計點的產品。
行動回報工作日誌、任務與下貨單，後台統計每日與每月營業額。
Pick one measurable workflow—e.g. support looking up documents, report summarization, or form field extraction. Define inputs, outputs, and human confirmation points first, then choose technology—don’t buy a model or build a chat box first.
Good fits are highly repetitive, have text/file inputs, and have controllable error cost. Irreversible actions involving money, contracts, or personal-data deletion should keep human confirmation and should not start fully automated.
Most enterprises start with commercial APIs. Building or fine-tuning models is costly and operationally heavy—usually only evaluated with large proprietary datasets, strict latency/offline needs, or when APIs can’t meet compliance.
General LLM APIs suit summarization, rewriting, and classification; company document Q&A usually needs RAG (retrieve then answer); Agents come in when you must query databases or call tools. Embeddings are the foundation for vectorizing data—often paired with RAG, not a standalone product.
AI PoC / MVP typically from NT$50,000; systems with knowledge bases, permissions, and a production admin backend usually cost more. Actual price depends on data prep, system integrations, and acceptance criteria—analysis gives a range, not a hard price.
PoCs commonly take 2–6 weeks to validate one scenario. After that we expand permissions, admin, monitoring, and launch. The usual path: requirement analysis → PoC → MVP → production system and operations.
This page explains how AI requirement analysis works and who it’s for. To book a paid slot for written architecture and cost judgment, see AI adoption consulting. If you’re ready to build a system, go directly to AI system development.
Share your current workflows and needs—we’ll help assess technical options, development scope, and estimated cost.
Book an AI requirement assessment