QuotePilot AI

• 

  QuotePilot_Workflow_summary

• 

  Qwen_extraction_tab

• 

  Tool_Calls_tab

• 

  PDF_+_OSS_upload_success

• 

  Human_Approval_tab

• 

  architecture_diagram

Inspiration

Businesses often receive messy quote requests through email or chat. A simple customer message can require several manual steps: understand the request, identify products, check inventory, calculate pricing, review risk, prepare a response, and create a quote document.

QuotePilot AI was inspired by this real business pain point. The goal was to build a Qwen-powered Autopilot Agent that can automate the RFQ-to-Quote workflow while still keeping humans in control of risky customer-facing actions.

What it does

QuotePilot AI turns customer inquiries into safe, human-approved quote workflows.

It can:

• Read messy B2B customer inquiries.
• Use Qwen Cloud to extract structured quote requirements.
• Detect missing information and ambiguous requests.
• Invoke deterministic business tools for catalog lookup, inventory checking, pricing, and policy review.
• Require human approval before customer-facing actions.
• Prepare an email draft without sending automatically.
• Generate a quote PDF after approval.
• Upload approved quote PDFs to Alibaba Cloud OSS.
• Run publicly from an Alibaba Cloud ECS deployment.

How we built it

QuotePilot AI was built with a practical agent architecture:

• Qwen Cloud handles customer inquiry understanding and structured extraction.
• Streamlit provides the web dashboard.
• Python business tools handle catalog lookup, inventory checks, pricing, policy guard logic, and email draft preparation.
• ReportLab generates quote PDFs.
• Alibaba Cloud OSS stores approved quote PDFs.
• Alibaba Cloud ECS hosts the deployed application.
• Docker and Docker Compose package and run the app.
• GitHub Codespaces was used for cloud-based development from a Samsung Galaxy Tab S8+.

The project separates AI reasoning from deterministic business logic. Qwen understands the messy customer message, while tool functions handle pricing, stock, policy decisions, PDF generation, and OSS upload.

Track 4 alignment: Autopilot Agent

QuotePilot AI is built for Track 4: Autopilot Agent.

It demonstrates:

• End-to-end business workflow automation.
• Ambiguous input handling.
• External tool invocation.
• Human-in-the-loop checkpoints.
• Production-readiness through secret handling, Docker deployment, audit records, cloud storage, and safety rules.

Human-in-the-loop safety

QuotePilot does not blindly automate risky business actions.

The policy guard requires review when:

• Required fields are missing.
• The inquiry is ambiguous.
• Inventory may be insufficient.
• The quote value is high.
• Urgent delivery or discount pressure is detected.
• A customer-facing response is being prepared.

The app can prepare drafts and generate PDFs, but it does not automatically send customer-facing messages without approval.

Challenges we faced

One major challenge was building and deploying the project from a tablet-only environment. The entire project was developed using GitHub Codespaces and a Samsung Galaxy Tab S8+.

Another challenge was designing a safe agent workflow. A normal chatbot could generate a quote-like answer, but that would be unsafe for real business use. QuotePilot needed deterministic tools, policy checks, and human approval to avoid hallucinated prices or unsafe commitments.

Deploying on Alibaba Cloud also required configuring ECS, Docker, security group rules, RAM permissions, OSS bucket access, and environment variables safely.

Accomplishments

We successfully built:

• A Qwen-powered inquiry extraction workflow.
• Business tool orchestration for catalog lookup, inventory, pricing, and policy guard logic.
• Human approval workflow.
• PDF quote generation.
• Alibaba OSS upload.
• Alibaba ECS deployment.
• Public app access through an ECS public IP.
• Documentation, architecture diagram, evaluation cases, Docker setup, and Devpost-ready proof assets.

What we learned

We learned that real AI agents need more than a model. They need:

• Tool use.
• State and workflow design.
• Safety policies.
• Human approval checkpoints.
• Clear audit trails.
• Cloud deployment readiness.
• Secure secret handling.

Qwen Cloud is powerful for understanding messy language, while deterministic tools make the business workflow more reliable and production-minded.

What's next

Next steps for QuotePilot AI include:

• CRM integration.
• Approved email sending.
• Role-based approval workflows.
• More product catalog examples.
• Automated evaluation metrics.
• HTTPS deployment with a domain.
• More advanced quote templates.
• Multi-user team workflows.

Built With

• aiagents
• alibabacloud
• alibabacloudecs
• alibabacloudoss
• b2b
• docker
• dockercompose
• githubcodespaces
• human-in-the-loop
• openai-compatibleapi
• python
• quoteautomation
• qwencloud
• reportlab
• streamlit
• workflowautomation