Which projects this form is for
- Technical knowledge search: give engineering, operations or after-sales teams answers with source references drawn from approved manuals, specifications and ticket records. See the technical knowledge solution.
- Support and ticket triage: classify incoming support requests, fill in missing information and prepare draft replies that people review before sending. See the support triage solution.
- RFQ intake: turn unstructured requests for quotation into structured requirements with missing fields flagged, while prices and commitments stay with your authorized team. See the RFQ intake solution.
- Other document-based workflows: if your case is not one of the three above, describe the document sources, the users and how you decide that an answer is correct.
Device, wholesale and general questions belong on the general contact form. CNPS is a FastGPT partner; platform licensing and implementation services are defined in a written proposal.
What happens next
- Clarify the scope: we reply to your work email to confirm the workflow, document sources, users and timeline. The first round is usually questions, not a proposal.
- Define the evaluation set: with your process owner, select a representative batch of questions or tickets and agree sample size, baseline and pass thresholds. A pilot without an evaluation set cannot produce a trustworthy conclusion.
- Written proposal: scope, licensing, data preparation, integration and support responsibilities, and the effort each side must contribute.
- Pilot and review: run against the evaluation set, compare results and decide whether to extend. People keep control of prices, availability, commitments and outbound replies.
We do not promise accuracy or efficiency figures before the pilot. Reference cases are published by FastGPT; CNPS has not independently audited their results.
What to prepare
- Document scope: which documents the assistant should answer from (type, volume, language, update frequency) and which are explicitly excluded.
- Users and setting: who asks, through which channel, roughly how many times a day, and the expected response style.
- How a correct answer is judged: who decides, against which criteria, and whether historical Q&A or tickets exist as a baseline.
- Systems and integration: where documents live, existing support or CRM systems, authentication method.
- Constraints: data-processing location requirements, access permissions, budget range and the timeline you have in mind.
- A process owner: one person who can decide on behalf of the business side.
Provide non-confidential examples only; do not upload or paste confidential documents into the form. See Define a document-grounded AI assistant and A knowledge AI pilot that earns trust.
Common questions
Our documents mix Chinese, English and Arabic. Is that workable?
It can be set as an evaluation goal. State the share of each language in the documents and the languages of the questions in the form; we include multilingual Q&A in the evaluation set rather than promising results in advance.
How long does a pilot take and how much of our time does it need?
It depends on the document scope and the size of the evaluation set. The written proposal lists the effort and time required from both sides; at minimum, a process owner and business staff who can judge answer correctness must take part.
Can FastGPT be deployed in our own environment?
Deployment options, licensing and data-processing location are defined in the written proposal. State data-residency or private-deployment requirements in the form and we will say plainly what can be met and what needs a separate assessment.
Can the figures in the reference cases be taken as our expectation?
No. Reference cases are published by FastGPT and CNPS has not independently audited them; your results depend on your documents, questions and criteria. Only the results of your own evaluation set serve as the basis for a decision.