BidPilot Qwen: Tender/RFP Autopilot Agent

Inspiration

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for BidPilot Qwen: Tender/RFP Autopilot Agent

Inspiration

Proposal teams spend days reading long RFPs, tenders, RFQs, and security questionnaires. A missed mandatory clause can disqualify a bid, and an unsupported compliance claim can create legal or customer trust risk. BidPilot Qwen is built around one rule: no evidence, no claim.

What it does

BidPilot Qwen ingests procurement documents and approved company evidence, extracts requirements, identifies deadlines and disqualification risks, searches for supporting evidence, drafts cited responses only when evidence exists, and routes risky items to accountable human reviewers.

The demo generates:

• requirements.json
• compliance_matrix.csv
• risk_summary.md
• draft_responses.md
• human_review_queue.md

How we built it

We adapted our existing Tender/RFP Compliance Copilot workflow to Qwen Cloud's Autopilot Agent track. The system is decomposed into an agent society: Intake Router, Requirement Extractor, Evidence Scout, Compliance Reasoner, Draft Writer, Human Review Gatekeeper, and Export Agent.

The repository includes a Python demo, deterministic validation logic, Qwen Cloud OpenAI-compatible API client, Alibaba Cloud Function Compute style backend, OSS deployment notes, schema, sample input, and generated outputs.

Qwen Cloud and Alibaba Cloud usage

The Qwen client calls the Qwen Cloud OpenAI-compatible Chat Completions API at:

https://dashscope-intl.aliyuncs.com/compatible-mode/v1

The Alibaba Cloud proof path includes:

• src/bidpilot_qwen/qwen_client.py
• src/bidpilot_qwen/alibaba_cloud_backend.py
• infra/alibaba-cloud/serverless-devs.yaml

Challenges we ran into

The hardest part was making the agent useful without encouraging hallucinated compliance claims. We solved this by separating extraction, evidence search, reasoning, drafting, and human review. Missing evidence on mandatory requirements becomes a blocker instead of a generated answer.

What we learned

Autopilot agents are strongest when they expose uncertainty and hand risky decisions back to humans. For proposal automation, the human-in-the-loop checkpoint is not a limitation; it is the product's safety boundary.

What's next

Next steps are deploying the Function Compute proof endpoint, adding PDF/DOCX/XLSX parsing, building a lightweight reviewer UI, and adding persistent evidence indexing with Alibaba Cloud OpenSearch or a vector database.

Built With

• ai-agents
• alibaba-cloud-function-compute
• alibaba-cloud-oss
• autopilot-agent
• human-in-the-loop
• openai-compatible-api
• procurement
• python
• qwen-cloud
• rfp