Qwen Cloud Track 4 Autopilot Agent

PrizePilot turns prize pages into accountable execution plans.

This submitted demo presents the same deterministic planning payload used by the CLI. It is designed for a fast judge review while keeping unverified sponsor evidence clearly separated from completed public artifacts.

Submission status: submitted_can_still_improve through Devpost. Public repo, Vimeo demo, and build journal are live; the project can still be improved until July 09, 2026 at 05:00pm EDT.

Current Qwen target: Blog Post Award - USD 500 x 10 winners.

Qwen Agent Flow

1. Public prize page

Input: Rules, awards, dates, required artifacts, and account-gated actions.

Qwen role: Frames the opportunity as a judge-facing story instead of a loose TODO list.

Proof: samples/qwen_hackathon.json and qwen-before-after evidence.

2. Deterministic rank

Input: Prize amount, winner count, deadline pressure, blockers, and evidence gaps.

Qwen role: Receives an auditable route plan rather than inventing the target from memory.

Proof: portfolio ranking in /api/plan and docs/api/plan.json.

3. Qwen refinement

Input: Blog Post Award route plus verified public artifact list.

Qwen role: Tightens the public narrative, risk wording, and update copy for judges.

Proof: live qwen-plus smoke receipt with non-sensitive output excerpt.

4. Approval gate

Input: Any secret, public posting, cloud billing, payout, tax, or KYC step.

Qwen role: Stops before irreversible account actions and records the human-controlled gate.

Proof: approval queue and integrity boundary in this dashboard.

5. Public proof packet

Input: Repo, Devpost, Vimeo, blog, judge hub, static plan JSON, and proof pages.

Qwen role: Turns the chosen route into a reviewable evidence trail without overclaiming.

Proof: judge pack, award evidence map, contribution map, and live proof page.

Qwen proof snapshot:

Verified: One Qwen/DashScope qwen-plus refinement pass ran with a runtime-only key and published non-sensitive evidence.
Public artifacts: Devpost, GitHub repository, Vimeo demo, Blog Award story, Pages hub, static judge demo, static plan JSON, Qwen contribution map, and Qwen live proof.
Still pending: Live Alibaba Cloud endpoint proof remains unclaimed until a public URL returns HTTP 200 at / and /api/plan.
Judge takeaway: PrizePilot is strongest as an honest autopilot: it accelerates route choice and submission evidence while keeping risky account actions visible.

Mind the Product target: First Place - USD 5000 cash. Novus status: email verification required.

Autopilot Walkthrough

Contest Analyst

Normalize prize pages into comparable opportunity records.

Deadline, prize, winner-count, blocker, and evidence fields.

Competitor Scout

Compare the route against public winning-project patterns.

Differentiators that judges can understand in under five minutes.

Differentiation Strategist

Choose the most reachable target instead of chasing the largest headline pool.

Blog Post Award / Honorable Mention path with explicit proof needs.

Submission Coach

Draft Devpost, blog, video, and deployment evidence assets.

Public links, scripts, runbooks, and update-ready copy.

Risk Reviewer

Hold account, billing, public posting, payout, tax, and KYC actions for human control.

Approval queue and evidence gap register.

Portfolio Ranking

#	Route	Target	Amount
1	Splunk Agentic Ops Hackathon	Most Valuable Feedback	USD 200
2	Global AI Hackathon Series with Qwen Cloud	Blog Post Award	USD 500
3	UiPath AgentHack	People's Choice	USD 500
4	Arm Create AI Optimization Challenge	Best in Category: Cloud AI	USD 1000

Judge Scorecard

Criterion	Weight	Proof	Gap
Innovation	30%	Autopilot for prize-route execution plus verified live Qwen refinement.	Turn the live proof into tighter public Devpost copy before judging.
Technical Depth	30%	Deterministic planner, Qwen-compatible client, HTTP API, Docker, Alibaba manifest.	Capture verified Alibaba Cloud endpoint proof.
Problem Value	25%	Turns real prize-route work into auditable actions and evidence gates.	Add more post-submission feedback if maintainers or judges request changes.
Presentation	15%	Public Devpost, Vimeo demo, blog, repo, and judge quickstart.	Refresh screenshots if the dashboard changes materially.

Trust and Control

Action	Status	Risk
Runtime Qwen/DashScope API key	completed once; future runs stay human-gated	secret
Alibaba Cloud live deployment	human-gated	billing/account
Devpost submission updates	allowed after review	public claim
Prize payout, tax, or KYC	user-only	financial identity

Evidence gaps before judging:

Qwen/DashScope live refinement is verified with a runtime key and no stored secret.
Alibaba Cloud Function Compute manifest exists, but no live endpoint has been verified.

Manual vs PrizePilot

Metric	Manual	PrizePilot
Manual prize triage	60-90 minutes	one structured plan run
Evidence tracking	spread across tabs and memory	status ledger plus dashboard
Risky actions	easy to mix with automation	explicit approval queue

Integrity Boundary

No API key is stored in the deployed app.
Public repository, video, blog post, and Devpost submission are recorded only after verification.
Live Qwen Cloud usage is captured as non-sensitive proof; Alibaba Cloud endpoint proof must still be captured before that claim is strengthened.
Novus is not claimed as installed until the official Novus account is verified and connected.

Machine-readable plan: /api/plan