prizepilot-qwen-cloud

Autopilot agent for ranking cash-prize hackathons, preparing submission assets, and gating risky external actions.

Comment

PrizePilot

PrizePilot is a Qwen Cloud Track 4 Autopilot Agent submission for turning hackathon and bounty opportunities into accountable execution plans. It ranks cash-prize routes, prepares public artifacts, and keeps sensitive external actions in a human approval queue.

Inspiration

Hackathon agents are usually judged by the app they produce. PrizePilot focuses on the work before the app exists: deciding which prize route is reachable, which evidence is missing, and which actions should stay gated because they affect accounts, public pages, cloud billing, or payout eligibility.

I built PrizePilot around a real portfolio: a near-term USD 200 feedback prize, the Qwen Blog Post Award / Honorable Mention route, and higher-friction open-source bounty backups. The goal is to make an autopilot agent useful without overclaiming evidence or pressing irreversible buttons blindly.

What it does

PrizePilot turns structured prize pages into:

  • ranked execution plans with deadline, prize, winner-count, and blocker scoring;
  • public artifact drafts for Devpost, blogs, videos, and deployment evidence;
  • a local dashboard and API for reviewing the current plan;
  • a judge-facing scorecard aligned to innovation, technical depth, problem value, and presentation;
  • a human action queue for sensitive steps such as API keys, OAuth, public uploads, cloud deployment, final submission, payment, tax, and KYC.

How it is built

The core planner is a deterministic Python package so every ranking can be audited locally. The web dashboard is a small standard-library HTTP service that exposes the same planning payload through / and /api/plan. A Dockerfile and Alibaba Cloud Function Compute manifest prepare the service for cloud deployment proof.

PrizePilot uses an OpenAI-compatible DashScope/Qwen client. A live Qwen/DashScope smoke test has verified that the local plan can be refined into sharper reasoning and submission narratives. The key is not stored in the repository or ledger, and future reruns still require action-time key entry.

Verified evidence

Qwen and Alibaba Cloud boundary

The Qwen/DashScope refinement path is verified with one runtime smoke proof using qwen-plus. The project still does not claim a live Alibaba Cloud endpoint. The repository includes an Alibaba Cloud Function Compute adapter and runbook, but endpoint proof should only be added after account-owner approval for any credit or billing use and after the endpoint is publicly reachable.

This boundary is a product feature: PrizePilot moves fast where preparation is reversible, and slows down where the action becomes public, costly, or account-bound.

What's next

The remaining evidence sprint is Alibaba Cloud endpoint proof: deploy or verify the dashboard on Alibaba Cloud, capture non-sensitive proof for / and /api/plan, then update this Devpost project only with proof that actually exists.

Built With

  • alibaba-cloud
  • autopilot-agent
  • dashscope
  • devpost
  • dockerfile
  • python
  • qwen
Share this project:

Updates