Nexural: a production platform I built and run by myself.

Next.js + React + Tailwind, custom JWT sessions, and Supabase Postgres 17. State-of-the-repo, not a template.

Real Stripe billing with signature-verified webhooks, transactional email, and web push — the parts most demos skip because they're hard to get right.

Refresh, ingest, scanner and nightly calibration engines pull from a commercial market-data API — with repo-level gates that fail the build if fake data sneaks in.

In-app AI features on the Vercel AI SDK, plus a separate always-on Discord engine running GPT-4o across dozens of automated market-cycle jobs.

• Row-level security on every one of 244 public tables
• Append-only admin audit log enforced by database triggers
• Zod-validated writes on the highest-risk admin paths
• Locked function search_path, env-gated admin 2FA

• A public track-record page that renders real engine output
• Build gates that block fake or stale data from shipping
• Revalidated every 5 minutes — not a screenshot from a good day

• 86 scheduled jobs keep the data and content fresh nightly
• Sentry error tracking + Vercel analytics wired in
• One person owns build, deploy, data, billing and support

A separate Node service — discord.js + GPT-4o + Supabase + Alpaca — deployed on Railway, posting market-cycle content and answering members across the server around the clock.

• 41 channels orchestrated
• 83 scheduled jobs on a market-hours cadence
• 76 database migrations, 299 test files

It's proof I can take an AI system past the demo — into something that runs unattended, on a schedule, wired to real brokerage and market data, and doesn't fall over.