ZhiKe AI

A Goal-Driven AI Business Agent for Salespeople and Client-facing Professionals

Dishui Lake Global OPC AI Challenge · S3 Global Youth Development Program · W3 Semifinal · X Innovation Track

ZhiKe AI helps salespeople, client managers, course consultants, business service consultants, and other client-facing professionals turn raw customer records into structured profiles, need analysis, opportunity assessment, follow-up plans, communication scripts, and daily business reports. In W3, it adds an in-session loop of business goals → confirmed follow-up feedback → KPI progress → next actions.

ZhiKe AI is neither a generic chatbot nor a data-storage-first CRM. It operationalizes customer-understanding and action-planning methods as an evaluable Skills workflow. KPI progress is driven only by user-confirmed business events; the model is not allowed to invent achieved results.

Core Agent Loop

Customer Input → Customer Profile → Need Analysis → Opportunity Assessment → Follow-up Plan → Communication Script → Daily Business Report → Follow-up Feedback → KPI and Action Adjustment

• Customer processing: Converts unstructured notes into traceable business outputs while separating facts, inferences, and unknowns.
• Feedback state: The user confirms events such as an effective conversation, need confirmation, completed demo, or priority-customer advancement.
• KPI action layer: Calculates progress from confirmed events, flags execution risks, and maintains an in-session priority action queue.

W3 Agent Demo

The current Streamlit application demonstrates a deliverable, interactive business Agent. It is not presented as a complete commercial system.

1. Set weekly or monthly business goals.
2. Paste customer notes and generate a seven-Skills business report.
3. Record a user-confirmed follow-up result in the KPI & Actions tab.
4. Review KPI progress, pace risks, and today's priority actions.

Live demo: https://zhike-ai-demo.streamlit.app/

Scope and Data Boundaries

• KPI metrics only count feedback confirmed by the user in the current browser session. AI suggestions are never treated as achieved performance.
• This version has no database, cross-session persistence, accounts, multi-user permissions, or live CRM, WeChat, or calendar integration.
• W2 mock customers remain limited to the cross-customer daily-report demonstration. W3 customer state and KPI data are separate, in-session demo data.
• Scripts, opportunity assessments, and business recommendations remain subject to human review and final decision-making.

Quick Start

cd zhike-ai
pip install -r requirements.txt
streamlit run app.py

W3 uses MiniMax as its default provider for report generation. When both MiniMax and SynScale are configured, MiniMax takes precedence and SynScale remains an optional fallback provider. Without a model key, or with Force Mock Mode enabled, the app runs the local Mock Skills Workflow for stable demonstrations and regression tests.

Local API configuration example. Never commit a real key to the repository:

MINIMAX_API_KEY=your_minimax_key
MINIMAX_BASE_URL=https://api.minimaxi.com/v1
MINIMAX_MODEL=MiniMax-M2.7

For Streamlit Community Cloud, add the same fields in App Settings → Secrets. Optional SynScale fallback fields are listed in .env.example; never commit a real key.

Project Structure

zhike-ai/
├── app.py                         # Streamlit W3 Agent Demo
├── README.md
├── README_EN.md
├── requirements.txt
├── skills/                        # Reviewable business Skill definitions
├── src/
│   ├── agent.py                   # Model providers and report fallback layer
│   ├── workflow.py                # W3 seven-step Skills orchestration and trace
│   ├── skills.py                  # W2 local Skills pipeline / Mock fallback
│   ├── kpi_agent.py               # W3 deterministic KPI and session action layer
│   ├── prompt.py
│   ├── schema.py
│   └── mock_customers.py
├── tests/
│   ├── test_workflow.py           # W3 Skills orchestration and fallback tests
│   └── test_kpi_agent.py          # W3 KPI regression tests
├── docs/
│   ├── 01_project_specs.md
│   ├── 02_skills_workflow.md
│   ├── 03_prototype_usage.md
│   ├── 04_demo_case.md
│   ├── 05_evaluation.md
│   ├── 06_roadmap.md
│   ├── 07_w3_agent_design.md
│   ├── 08_kpi_framework.md
│   ├── 09_w3_demo_script.md
│   ├── 10_w3_evaluation.md
│   ├── 11_w3_test_evidence.md
│   └── evidence/                    # Reviewable W3 runtime evidence
│       ├── w3_api_case_01_training.png
│       ├── w3_api_case_02_course_consultant.png
│       ├── w3_api_case_03_enterprise_service.png
│       ├── w3_kpi_feedback_interface.png
│       └── w3_kpi_feedback_confirmed.png
└── prototype/                     # W2 reference interaction page

Verification

python tests/test_kpi_agent.py
python tests/test_workflow.py

The regression tests confirm that repeated generation does not duplicate a customer, only user-confirmed feedback changes KPI values, and the KPI layer does not require an external model call.

Documents

• Project Specs
• Core Skills / Workflow
• W3 Agent Design
• KPI Framework
• W3 Demo Script
• W3 Evaluation
• W3 Test Evidence
• Chinese README

Competition Information

• Current stage: S3 Global Youth Development Program · W3 Semifinal
• Track: X Innovation Track
• W3 task: Deliver a runnable Agent that integrates Skills and demonstrates interaction and user value.
• Frozen W2 baseline: Git tag w2-final, preserving the W2 prototype version.