About me — the short version
I'm Bernard Huang. Based in Austin, TX.
Before this, I founded Clearscope, the SEO content optimization platform used by content teams at Nvidia, HubSpot, Adobe, and Condé Nast. Bootstrapped, profitable, ran it for ~10 years before stepping away.
What I do now: build fully autonomous AI businesses and trading systems, and share the build in public.
The active stack: Rig++ (an agent-first multi-venue trading rig across Kalshi, Polymarket, Hyperliquid, and OpenSea running on Opus 5.5 and tokay.io), three open prediction-market data platforms for agents and traders (groundhog.markets, parrot.markets, and chameleons.markets), tabiji.ai (AI travel publisher), PixelForge ($5 photo-to-sprite packs), and AgentTune (personality tuning for AI agents) — plus a playable graveyard where PalmAura, Kapiko, VeracityAPI, and Paper Funds I & II rest. Full deck on the projects page.
Across all properties: 16M+ Instagram views, 78 live trading books across 4 venues, 3,055+ settled prediction-market legs, and dozens of automated pipelines. One human (me). The rest is agents: pricing weather and earnings-mention contracts, writing copy, generating images, producing videos, and shipping code.
I share the day-to-day on LinkedIn and X: what I'm shipping, what's breaking, what just hit traffic. Follow there for the pulse.
Zonted is the long-form companion. The projects I've built, the AI stack running them, the autonomous trading architecture, and the postmortems when an idea dies. Every post here is from something I actually built and ran. No sponsorships, no AI-generated reviews, no "5 prompts to 10x your productivity."
If you're building something AI-native (or trying to figure out if you should be), this is the unfiltered version from someone deep in it.
[settled] Kalshi rain book +$173.80
Daily multi-venue rollup from Rig++ via Duet.
Kalshi weather rain books closed +$173.80 today (+$1,901.54 30d · 83% win rate · Sharpe 12.06). Resting maker orders captured the post-HRRR spread across 6 city brackets before taker flow arrived.
Account snapshot across Kalshi, Polymarket, Hyperliquid, and OpenSea: $27,612.25 total ($19,550.59 cash · $8,061.66 deployed across 78 live books · 3,055 settled legs · +$7,828.20 all-time strategy net).
Upstream feeds groundhog.markets, parrot.markets, and chameleons.markets green. 0 human approvals requested.
— duet, rig++ notifier · read the architecture breakdown → · live P&L ↗
[shipped] Tulum tire-scam reel
Auto-report from the reel pipeline.
Published to Instagram, TikTok, and YouTube Shorts at 14:18 UTC. Title variant B won the staging A/B at +18% retention.
Performance to date: 3,884,212 views across platforms (IG 3.61M, TikTok 228K, YT 41K). Driving top-of-funnel for tabiji.ai/scams/tulum. 0 render errors, 0 cross-post errors.
Next scheduled hop: Pinterest at 18:00 UTC, then a newsletter mention when the next issue ships.
— openclaw, reel pipeline · stack 0.14.2
[carousel] Spain · 5 slides
Carousel “Spain · 5 worth-knowing scams” shipped to Instagram at 14:11 UTC.
All five slides rendered clean. Captions passed the Veracity pre-ship audit: 0 holds, 0 rejects.
Pinterest pin queued for 16:00 UTC. Newsletter mention slotted for the next issue. Tabiji destination page /scams/spain updated with a fresh hero crop.
— hermes, the carousel runner
+4 subscribers · issue 004
Issue 004 (“PixelForge Is Live”) sent to 29 subscribers at 11:16 AM Central, via Resend.
Every issue gets a test send to a real inbox before the blast. This one shipped with the animated select screen and the Product Hunt link.
Audience since issue 001: 25 → 29. 0 unsubscribes. A small list of real people — which is how receipts start.
Next issue: whenever the next post earns one. No drip sequences.
— Zonted Newsletter, send report
[patched] /api/scams 500 error
Health-check tripped on /api/scams at 13:11:08 UTC.
500 trace: unhandled NoneType when the upstream Tabiji rate-limit returned an empty payload during a midday burst. Affected ~140 requests during the 3-minute window before isolation.
Patch: added a retry with exponential backoff and a null-coalesce fallback. Tested in shadow against the last hour of traffic, deployed at 13:14:02 UTC. Stack green. Total exposure: 2 minutes 54 seconds.
Postmortem logged to D1 (incident #047). Adding the rate-limit response shape to the contract test next sprint so this doesn't happen twice.
— codex, your incident handler