Service Description
1.1 Service Overview
LLM Consulting Service:
Scenario-based consulting for intelligent computing needs, providing technical consulting and guidance services based on the Tongyi LLM series products, with on-demand prompt optimization, model evaluation support, and other technical services.
AI Agent Development Technical Service:
Providing AI agent construction services based on client business scenarios, including business scenario analysis, agent architecture design, engineering pipeline implementation, and performance monitoring & optimization for end-to-end agent delivery.
Model Fine-tuning & Training Service:
Providing domain knowledge adaptation and model customization for enterprise clients, delivering end-to-end technical services from data preparation and model selection to training and tuning; supporting fine-tuning and training methods including SFT, LoRA, DPO, CPT, etc., with complementary services including training data cleaning and annotation, training process monitoring, model performance verification, and model deployment.
Service Scope
2.1 LLM Consulting Service Scope
Service scope includes:
Scenario requirement research: Interviews to confirm business pain points and LLM application requirements, research on existing knowledge bases, data assets, API interfaces, and other resources.
Technical status research: Analyze computing resources, network environment, and technical architecture; clarify the integration capabilities and interface specifications of knowledge sources, databases, and application systems.
Agent orchestration solution design: Design agent pipelines, nodes, model selection, evaluation methods, and standards.
Context engineering optimization plan: Provide scenario-oriented in-depth prompt optimization services.
Project process management and consulting plan development support.
Service scope does not include:
Engineering development related to AI agents.
Code development and functional modifications of client systems.
Daily O&M services for third-party software including installation, testing, fault diagnosis, and optimization.
IT system architecture transformation or general digital transformation consulting unrelated to LLMs.
2.2 AI Agent Development Service Scope
Service scope includes:
LLM-related application scenario requirements and technical research analysis.
Agent architecture design and orchestration implementation.
Context engineering design and implementation.
Agent performance evaluation and optimization. (Excluding model fine-tuning and training)
Tool or Skills integration.
Project process management and implementation plan development support.
Current AI agent development scenario scope includes:
No. | AI Agent Scenario | AI Agent Scenario Description |
1 | Complex Document Knowledge Base RAG Q&A | Help enterprises efficiently retrieve massive unstructured complex documents within the organization, support natural language queries, automatically retrieve document segments, generate precise answers with context, significantly reduce knowledge search costs, and improve decision-making efficiency. |
2 | Form Information Extraction & Filling | Help enterprises quickly process various business forms, automatically extract key fields from uploaded images/documents, intelligently verify logical completeness, and pre-fill target system forms, greatly reducing manual input errors and time consumption. |
3 | Complex Structured Report Generation | Help enterprises efficiently generate professional structured reports. Based on input objectives/data/templates, automatically organize content frameworks, populate analytical conclusions, and support multi-version comparison and compliance verification, significantly improving report quality and efficiency. |
4 | Airline & Hotel Travel Assistant | Help aviation and travel enterprises build full-lifecycle travel assistants. Based on traveler preferences, budgets, etc., integrate the enterprise's product resources to generate detailed travel plans, support in-trip fulfillment management and emergency plan recommendations, enhancing the traveler experience. |
5 | Smart 3C & Home Appliance Manager | Help 3C enterprises build smart home appliance assistants. Based on ASR+LLM+TTS pipeline adapted for all consumer electronics devices, achieve 'see-and-speak' zero-layer interaction, significantly enhancing smart home appliance user experience and home intelligence level. |
6 | HR Intelligent Recruitment | Help enterprises efficiently complete the full recruitment process from job posting to candidate onboarding, automatically parse JDs, distribute across multiple channels, intelligently screen resumes, generate evaluation reports, and support candidate experience tracking, significantly shortening recruitment cycles. |
7 | Person-Job Matching & Training Planning | Help enterprises accurately assess the match between employee capabilities and job requirements. Based on employee history, performance, skills and other tags, automatically identify capability gaps, recommend learning paths and training courses, supporting talent development and organizational effectiveness improvement. |
8 | Logistics Route Code Governance & Delivery Verification | Help logistics enterprises efficiently handle address standardization and delivery verification in last-mile delivery. Automatically parse ambiguous addresses, match standard route codes, support delivery photo/signature recognition and anomaly detection. |
9 | Public Sentiment & VOC Analysis | Help enterprises monitor and analyze customer feedback and voice across social media, reviews, and other channels. Identify sentiment trends, hot topics, and risk signals, generate insight reports and recommend response strategies. |
10 | Insurance Claims One-stop Assistant | Help insurance enterprises automate claims processing, including parsing claim materials, comparing policy terms, assessing loss reasonability, identifying fraud risks, generating claim conclusions and facilitating communication, improving claims efficiency. |
11 | Product Tagging & Selection Assistant | Help e-commerce enterprises with product tagging and selection tools. Parse product information, match category tags, identify competitors, and recommend high-potential products based on sales data, improving selection accuracy. |
12 | Online Virtual Medical Assistant | Help medical enterprises build virtual medical assistants. Through multi-turn conversations to collect symptom information, combined with knowledge bases to provide cause analysis and self-check suggestions, supporting health consultation, appointment booking, and user follow-up. |
13 | Medical Record Classification & Tagging | Help medical enterprises efficiently process medical records, parse clinical text, identify key entities such as diseases, medications, etc., classify and tag according to standard systems, supporting quality control management and medical research. |
14 | Smart PPT Copywriting Assistant | Help enterprise users generate PPT copy efficiently (including outlines, drafts), automatically generate logical page layouts, page copy, and chart suggestions, supporting style adaptation and template configuration. |
15 | Intelligent Survey Assistant | Help research enterprises generate and distribute questionnaires. Based on research objectives, recommend question templates, select channels, etc., generate personalized questionnaires, collect and analyze responses, generate insight conclusions and visualization reports. |
16 | Intelligent Bidding Assistant | Help enterprises efficiently handle bidding documents. Support tender document summarization, bid document auto-generation, bid evaluation criteria review, bid resource matching, and intelligent bid response review. |
17 | Restaurant Smart Ordering Manager | Help catering enterprises build ordering systems. Support customer natural language ordering, recommend based on order history, connect to product knowledge base and inventory, support integration with ordering systems to achieve end-to-end ordering workflow. |
18 | Enterprise BI Data Query & Analysis Assistant | Connect to enterprise internal databases, support natural language queries to generate SQL queries and data insights, combined with client business domain knowledge base to provide data analysis suggestions. |
Service scope does not include:
Code development and functional modifications of client systems.
Frontend application development and functional modifications of AI agents.
Daily O&M services for third-party software including installation, testing, fault diagnosis, and optimization.
IT system architecture transformation or general digital transformation consulting unrelated to LLMs.
2.3 Model Fine-tuning & Training Service
Service scope includes:
Scenario requirement research for fine-tuning/training.
Fine-tuning/training solution design.
Dataset preparation (including training data and evaluation data).
Performance evaluation and optimization.
Model deployment and launch support.
Project process management and implementation plan development support.
Service scope does not include:
Code development and functional modifications of client systems.
Installation, testing, fault diagnosis, and optimization of third-party closed-source software.
Training/inference frameworks beyond Alibaba Cloud products, and all framework-level modifications (including both Alibaba Cloud and third-party frameworks).
Prerequisites
3.1 AI Consulting Service Prerequisites
Service request timeline: Client must submit the service request at least 15 calendar days in advance;
Client provides business scenario description, existing system architecture documentation, and knowledge base/data asset inventory.
Client must purchase and use Alibaba Cloud Tongyi series products. (At minimum, the Tongyi model)
Environment & access: Client must provide non-production environment access, remote channels, and necessary data/code resources.
Project cooperation: Client must designate a project manager or technical manager with decision-making authority as the key contact, and provide written confirmation of the implementation plan within 5 business days.
3.2 AI Agent Development Service Prerequisites
Service request timeline: Client must submit the service request at least 15 calendar days in advance;
Environment & access: Client must provide non-production environment access, remote channels, and necessary data/code resources.
Client must purchase Alibaba Cloud computing resources or use Alibaba Cloud MaaS platform services.
Project cooperation: Client must designate a project manager or technical manager with decision-making authority as the key contact, and provide written confirmation of the implementation plan within 5 business days.
3.3 Model Training Service Prerequisites
Service request timeline: Client must submit the service request at least 15 calendar days in advance; for scenarios involving large-scale data or computing resource migration, 30 days advance notice is recommended.
Client should provide necessary non-production environment, data samples, API interface documentation, and remote access permissions.
Client provides training datasets, computing resources, clear task objectives, and evaluation criteria.
Project cooperation: Client must designate a project manager or technical manager with decision-making authority as the key contact, and provide written confirmation of the implementation plan within 5 business days.
Client must purchase Alibaba Cloud computing resources or use Alibaba Cloud MaaS/Intelligent Computing platform services
Alibaba Cloud does not provide reinforcement learning reward function design; the reward function must be provided by the client.
Responsibility Boundaries & SLA
4.1 Responsibility Boundaries
4.1.1 Client & Alibaba Cloud
Client purchases LLM technical services.
Both parties negotiate and confirm the specific business objectives and scope for this service.
4.1.2 Client
Define business objectives.
Provide facilities, equipment, necessary non-production environments, remote access channels, and permissions to support Alibaba Cloud in delivering services.
Cooperate with Alibaba Cloud in researching existing technical architecture, model training configurations, computing resource utilization, etc., and participate in specific plan implementation such as deployment plans, migration plan development, etc.
Review the implementation plan developed by Alibaba Cloud, and confirm it in writing or via email. The client shall not reject Alibaba Cloud's technical recommendations or solutions without legitimate technical justification.
Execute specific migration, deployment, and optimization plans.
As the O&M entity, be responsible for relevant O&M work.
4.1.3 Alibaba Cloud
Responsible for project organization and establishing an expert solution implementation team.
Understand client business objectives and scope, develop implementation plans, and obtain written (including but not limited to email) confirmation from the client.
Provide all service items specified in this work description, such as business objectives, agent development plans, model evaluation plans, model optimization plans, etc., and provide feasible recommendations.
4.1.4 Completion Standards
Submit the "Service Acceptance Report" and obtain client acceptance approval.
4.2 SLA
Alibaba Cloud shall provide a dedicated LLM service technical manager.
Alibaba Cloud shall provide the "LLM Technical Service Work Plan" and the "LLM Technical Service Acceptance Report"
Service Items
5.1 AI Consulting Service Items
Each package provides research and consulting services including:
Requirement scenario research, providing LLM scenario research and analysis services, including:
Interviews to confirm client business pain points and LLM application requirements.
Research on existing knowledge bases, data assets, API interfaces, and other available resources.
Each package supports no more than 3 research meeting sessions per business scenario.
Technical status research, providing LLM-related technical status research and analysis services, including:
Analyze the client's current computing resources, network environment, technical architecture, and other existing conditions.
Clarify the integration capabilities and interface specifications of enterprise knowledge sources, databases, and application systems.
Each package supports technical integration assessment for no more than 5 major systems.
Agent technical pipeline design, including:
Output agent-related technical solutions, including agent orchestration pipelines, node design, model selection, evaluation methods and standards, etc.
Design agent engineering pipelines, develop multi-node workflow or multi-agent workflow solutions for complex scenarios, and provide technical consulting guidance for agent development.
Each package supports 1 pipeline design solution with no more than 5 nodes.
Scenario-oriented prompt optimization services, including:
Tongyi Qwen model prompt compatibility rewriting.
Prompt performance evaluation and PE tuning.
Each package supports no more than 3 rounds of prompt optimization, with no more than 5 models and no more than 1,000 evaluation data records per round.
Provide project management services throughout the project execution process.
5.2 AI Agent Development Service Items
AI agent development technical services include:
Scenario requirement research, providing LLM scenario research and analysis services, including:
Interviews to confirm client business pain points and LLM application requirements.
Research on existing knowledge bases, data assets, API interfaces, and other available resources.
Each package supports no more than 3 research meeting sessions per business scenario.
Technical status research, providing LLM-related technical status research and analysis services, including:
Analyze the client's current computing resources, network environment, technical architecture, and other existing conditions.
Clarify the integration capabilities and interface specifications of enterprise knowledge sources, databases, and application systems.
Each package supports technical integration assessment for no more than 5 major systems.
Agent technical pipeline design and implementation, including:
Output agent-related technical solutions, including agent orchestration pipelines, node design, model selection, evaluation methods and standards, etc.
Design MCP tools or Skills required by the agent engineering pipeline, and define the input/output standards for the tools.
Each package supports no more than 1 orchestration pipeline, 5 orchestration nodes, and 3 MCP tools/Skills configurations.
4) Context engineering optimization, including:
Design prompts for nodes in the orchestration solution and conduct multi-round optimization.
Based on Alibaba Cloud MaaS products, combined with the agent orchestration solution and context engineering, conduct engineering pipeline integration and implementation, and complete integration with the client's existing tools or Skills.
Each package supports no more than 5 rounds of prompt optimization, with optimization verification for no more than 5 models.
5) Agent evaluation verification and performance optimization, including:
a) Provide agent performance verification, provide high-availability deployment solutions for agents, and support clients in completing related deployment work according to the deployment plan.
b) Provide bad case collection and optimization services, and conduct performance tuning based on agent architecture without model fine-tuning. The final delivered results shall be subject to the standards jointly confirmed by both Alibaba Cloud and the client.
c) Each package supports no more than 2,000 evaluation data records (bad case collection and optimization no more than 5 rounds), with evaluation and optimization verification for no more than 5 models.
Provide project management services throughout the project execution process.
5.3 Model Fine-tuning & Training Service Items
Model fine-tuning and training services include:
Scenario requirement research, providing LLM fine-tuning or training scenario research and analysis services, including:
a) Interviews to confirm client business pain points and LLM fine-tuning/training requirements.
b) Research on existing training data, annotation resources, computing budget, and other available resources.
c) Each package supports no more than 3 research meeting sessions per fine-tuning/training scenario.Data collection and computing resource assessment, providing client-side data quality collection analysis and computing resource assessment services, including:
a) Collect and analyze client's current data quality and data format, and provide data format specifications and sample quality inspection recommendations. (Datasets to be provided by the client)
b) Assess the specifications and quantity of resources used for fine-tuning/training.
c) Perform data preprocessing on collected client data as needed, including data deduplication, filtering, augmentation, etc.
d) Each package supports technical integration assessment for no more than 5 major data sources, with client-provided data volume not exceeding 50,000 samples.
Fine-tuning/training solution design and training task execution, including:
a) Output fine-tuning/training related technical solutions, including base model selection, fine-tuning/training strategies (e.g., LoRA/SFT/RLHF), hyperparameter design and evaluation criteria, etc.
b) Configure training scripts and execute training tasks (based on computing resources provided by the client).
c) Each package supports no more than 1 training task + 1 model version + 1 algorithm, with no more than 3 model training iterations.Model evaluation verification and performance optimization, including:
a) Design core metrics based on client evaluation objectives, such as accuracy, relevance, compliance, and response completeness, etc.
b) Execute model evaluation and deployment, output model evaluation report. (Evaluation dataset to be provided by the client)
c) Provide bad case collection and optimization services, and conduct performance tuning based on fine-tuning/training architecture. The final delivered results shall be subject to the standards jointly confirmed by both Alibaba Cloud and the client.
d) Each package supports no more than 50,000 evaluation data records, no more than 3 evaluation iterations, and evaluation and optimization verification for no more than 3 model versions.
Provide project management services throughout the project execution process.
Service Workflow
6.1 LLM Consulting Service
Application deadline: Client must submit the application at least 15 calendar days before the LLM consulting service start date.
6.2 AI Agent Development Service
Application deadline: Client must submit the application at least 15 calendar days before the AI agent development service start date.
6.3 Model Fine-tuning & Training Service
Application deadline: Client must submit the application at least 15 calendar days before the model fine-tuning & training service start date.
Acceptance Criteria
Alibaba Cloud provides the following deliverables, which constitute qualified service acceptance:
Alibaba Cloud delivers the "LLM Technical Service Work Plan" and the "LLM Technical Service Acceptance Report", and obtains written confirmation from the client (including email format).
The "LLM Technical Service Work Plan" and the "LLM Technical Service Acceptance Report" include the following content:
Prior to service commencement, Alibaba Cloud's analysis of client LLM requirements and LLM technology selection recommendations.
Based on the client's business characteristics and requirements, Alibaba Cloud provides technical solution recommendations for client business implementation.
Completion Standards
Implementation completed and client acceptance finalized.