Entry Gate

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 Entry Gate

What it does

Entry Gate is a simulated Alexa+ submission steward. Instead of answering trivia about contest rules, it verifies the artifacts you are about to submit.

You paste a live demo URL, a GitHub repo, and (optionally) a YouTube or Vimeo link, then ask whether you are ready to submit. Entry Gate runs a short tool loop:

1. fetch_page — GET the demo; report HTTP status, title, and soft failures (auth walls, thin pages).
   
2. check_repo — confirm the repo is public, has a LICENSE and README (with light setup-quality hints), description, and recent push.
   
3. check_video_meta — oEmbed title/duration when a video URL is present; flag videos over 3 minutes.
   
4. score_checklist — deterministic rubric → go / almost / no-go, score, ranked fixes, plus Amazon product-feedback / friction-log reminders.
   
5. Optional MiniMax narration for a spoken-style summary (falls back to rules if the key is unavailable). The UI also offers browser Speak (Web Speech API) and keeps URLs in session storage across turns — multi-step state, not a single-shot FAQ.

Why this for the Alexa+ track: Amazon’s creative guidance favors multi-step orchestration and state across a session. The product is verification — “don’t trust the agent’s claim that work is done.” The dark conversational panel is meant to feel like chatting with an Alexa+-style assistant, without requiring a physical Echo or a live MCP hook to Alexa.

Self-check: auditing Encore (encore-rouge-chi.vercel.app + qinheyi/encore) returns almost with a clear “record a <3 min demo” fix when no video is provided — that is the intended judge path. Use the Try Encore and Audit Entry Gate chips on the live demo.

Out of scope on purpose: no Fire TV / Bee / Ring hardware, no Echo skill certification — the rules allow a simulated Alexa+ experience with any agentic stack.

Inspiration

Hackathon submissions fail for boring reasons: dead demos, private repos, missing licenses, overlong videos. Entry Gate turns those checks into one conversational audit so a builder (or an agent) can fix the list before Submit.

How we built it

Next.js + TypeScript on Vercel, a tool loop in /api/audit (fetch_page, check_repo, check_video_meta, score_checklist), the public GitHub REST API, YouTube/Vimeo oEmbed, optional MiniMax narration, and the browser Web Speech API.

Challenges we ran into

Deterministic scoring had to stay trustworthy when the LLM is offline; MiniMax narration is optional. GitHub checks use the public HTTP API so the deployed demo works without a server-side gh login. YouTube oEmbed omits duration, so the checklist must warn instead of guessing.

Accomplishments that we're proud of

• Live demo + public MIT repo
  
• Tool-verified audit path (demo soft-fail sniff, repo README quality, video, checklist)
  
• Alexa+-style UX: session memory, example chips, Speak, follow-up prompts
  
• Works without MiniMax; better spoken summary with it

What's next for Entry Gate

Ship the demo video; paste product feedback + friction logs on Devpost; optional contest rule packs beyond Devpost-shaped defaults; optional Bedrock/AgentCore if free-tier credits appear.

Built With

• alexa+
• github-api
• minimax
• next.js
• typescript
• vercel
• web-speech-api