gpu/single-gpu/home_benjamin_hdbasin/code/docs/public_demo_setup.md

HD-Basin Public Demo Setup

Target URL: https://hdbasin.pages.dev

Deployment status as of 2026-07-01:

• Cloudflare Pages project hdbasin exists.
• Production URL https://hdbasin.pages.dev returns HTTP 200.
• API health route https://hdbasin.pages.dev/api/health returns OK.
• Supabase env vars are configured in Cloudflare Pages.
• Free demo route POST /api/demo/session is available for no-payment queue tests.
• Stripe env vars are still pending until the Stripe account/test keys exist.

This repo now has the local foundation for the public demo:

• Vite React frontend in web/
• Cloudflare Pages Functions under web/functions/api/
• Supabase migration under supabase/migrations/
• Stripe Checkout + webhook API routes
• A100 worker queue harness at scripts/a100_worker.py

Accounts

Create or have access to:

1. GitHub
2. Cloudflare
3. Supabase
4. Stripe
5. Gmail or other support email
6. The current private A100 box

Optional later:

• Runpod or Modal for pay-on-demand GPU workers
• Sentry for error tracking
• UptimeRobot or Better Stack for uptime checks
• PostHog for funnel analytics

Cloudflare Pages

Sign up: https://dash.cloudflare.com/sign-up

Create a Pages project connected to the GitHub repo.

Use:

Project name: hdbasin
Production URL: https://hdbasin.pages.dev
Root directory: web
Build command: npm run build
Build output directory: dist

Set these environment variables in Cloudflare Pages:

SUPABASE_URL
SUPABASE_SERVICE_ROLE_KEY
STRIPE_SECRET_KEY
STRIPE_WEBHOOK_SECRET
PUBLIC_APP_URL=https://hdbasin.pages.dev

Supabase

Create a Supabase project and run:

-- Paste and run:
-- supabase/migrations/202607010001_hdbasin_public_demo.sql

The migration creates:

• public.hdbasin_jobs
• public.hdbasin_job_events
• queue RPC functions for the worker
• private storage buckets hdbasin-uploads and hdbasin-results

Use the service-role key only in Cloudflare Pages Functions and the private A100 worker. Do not expose it in frontend code.

Stripe

Start in test mode.

Use Stripe Checkout Sessions for one-time payments.

Webhook endpoint:

https://hdbasin.pages.dev/api/stripe/webhook

Minimum events:

checkout.session.completed
checkout.session.expired
payment_intent.payment_failed

The webhook is the only thing that moves a job into queued.

Local Cloudflare Preview

cd web
cp .dev.vars.example .dev.vars
# Fill .dev.vars.
npm run pages:dev

Health check:

curl http://127.0.0.1:8788/api/health

A100 Worker

On the private GPU box:

cp .env.a100-worker.example .env.a100-worker
# Fill SUPABASE_URL and SUPABASE_SERVICE_ROLE_KEY.
set -a
. ./.env.a100-worker
set +a
python3 scripts/a100_worker.py --once

Without HDBASIN_RUN_COMMAND, the worker writes a pipeline-only demo report. For the public demo, the A100 worker is now wired to a real small AG News DistilBERT run through scripts/run_public_demo_job.py.

Once the paid runner is ready, set:

HDBASIN_RUN_COMMAND="python3 scripts/run_public_demo_job.py --job-id {job_id} --outdir {outdir} --trial-budget {trial_budget} --workload {workload} --file-name {file_name}"

This still uses a built-in public benchmark workload. Replace it with the real customer-dataset runner after private dataset upload/download is wired.

Demo Gate

Before public traffic:

1. Cloudflare Pages deploy loads at https://hdbasin.pages.dev.
2. /api/health returns OK.
3. Supabase migration is applied.
4. Stripe test webhook succeeds.
5. Test checkout creates a job.
6. Webhook changes the job to queued.
7. A100 worker claims the job.
8. Worker uploads report artifacts.
9. Job ends as succeeded.
10. Failure path marks jobs as failed and does not leave them running forever.