How Tianyi Cloud Shifts from Manual Ops to Model‑Driven AI+ Cloud Integration

AI+ and Cloud Service Acceptance Challenges

Cloud service acceptance involves a large product catalog, flexible configurations, multiple pricing models, and strict compliance clauses, which create high entry barriers for non‑technical users.

Product system is massive; users must match requirements to instances (general, memory‑optimized, HPC) and configure CPU, memory, storage, bandwidth, etc.

Pricing models include pay‑as‑you‑go, reserved instances, annual/monthly packages; discount rules differ, making cost prediction difficult.

Contracts contain SLA, data‑sovereignty, privacy, liability clauses that require careful interpretation.

Traditional “Customer‑Manager” Model

Most providers assign a customer manager to perform pre‑sales consulting and order placement, which shifts operational burden to the provider without solving product selection, pricing comprehension, or contract opacity.

Intelligent Assistant Solution

An AI‑driven “Intelligent Assistant” built on a large language model and natural‑language interaction is proposed with five objectives:

Personalized service for each manager.

100 % self‑service handling.

Workflow automation.

Conversational business processing.

Optimal deployment and cost‑estimation recommendations.

Technical Architecture

The solution consists of three tightly coupled platforms:

Data middle‑platform : ingests business data (orders, resource status, account information, knowledge‑base documents), performs cleaning, desensitization, and stores structured training sets.

AI middle‑platform : trains models that interpret user intent and generate compliant solutions.

Business middle‑platform : exposes APIs that invoke AI models, replacing manual clicks with model‑driven guidance.

Figure 1 illustrates the architecture.

Model Development Strategy

The large model is developed in‑house to satisfy a data‑security strategy. Training requires access to sensitive configuration and account data; a private on‑premise training environment prevents data leakage and allows rapid model iteration aligned with evolving business needs.

Core Assistant Functions

Personal dashboard : shows pending approvals, expiring resources, applicable discounts, and price‑drop alerts.

Resource renewal : user provides a resource ID; the system validates the ID, enforces fine‑grained permission checks (e.g., manager A can renew only his customers’ resources), and executes renewal automatically, eliminating manual errors. Figure 3 demonstrates the renewal flow.

Configuration recommendation : based on user requirements, budget, and workload, the assistant generates a recommended product configuration and a preliminary cost estimate. The recommendation is advisory but reduces the need to consult multiple product documents. Figure 4 shows an example.

Operational diagnostics : operators can upload error logs or screenshots; the assistant triggers an initial diagnosis workflow using the same AI capabilities. Figure 5 illustrates this use case.

Continuous Model Training

Model updates draw from business‑platform artifacts such as requirement specifications, detailed design documents, and SOPs uploaded via management portals. After nearly two years of iteration, the assistant achieves:

100 % coverage of business scenarios.

Indexing of over 500 knowledge‑base documents.

Handling 70 % of daily order volume.

Overall work‑efficiency improvement of 50 %.

The development team aims to raise coverage to 90 % by further enriching the knowledge base and adding functional modules.

Code example

“IT有得聊”是机械工业出版社旗下IT专业资讯和服务平台，致力于帮助读者在广义的IT领域里，掌握更专业、更实用的知识与技能，快速提升职场竞争力。 点击蓝色微信名可快速关注我们!