a hobby crypto bot I built fades @HyperliquidX order book imbalance — when bid walls stack 10x heavier than asks, it shorts the wall.
theory: visible walls invite front-running.
watch the algorithms work (2hr delayed)
https://t.co/0nY4Enlewv
#Hyperliquid#algotrading
SOL 4H signal stack from my hobby algo bot.
Today: shipped Donchian shadow exits + slip-adjusted P&L so the next 30 days of data shows what would *actually* fill — not what looks clean in hindsight.
Hobby project. Not advice. Not registered SEC/CFTC/FINRA.
#algotrading #typescript #hyperliquid #buildinpublic
4/4
now in visual-validation mode before adding the next agent. one agent. validate. expand. that's the only thing that separates this from "i built 12 agents that produce 12 contradictory signals"
next: flow (rvol, cvd), then funding/OI. LLM agents last
#Perps#CryptoTwitter #Quant
3/4
discipline that makes it work:
• canonical feature registry. no agent invents its own vocabulary
• idempotent writes, bar_ts ≠ computed_at, ADRs for schema changes
first agent: 1h structure across BTC ETH SOL HYPE ARB. ~1k observations, 16 BOS/CHoCH events. smoke tests green
#BuildInPublic
2/4
architecture:
• timescale as the substrate
• containerized agents, one per timeframe × feature family
• agents observe, they don't decide. directional logic stays in the strategy layer
• every signal logged with full feature snapshot. db becomes the research env
#AIAgents #DeFi
1/4
shipped v0 of a multi-agent observation layer for hyperliquid perps tonight
the edge isn't "LLM predicts price." it's breadth of attention across 150 markets, structured into a db your strategies can query for confluence before executing
#Hyperliquid#AlgoTrading
counter-intuitive: knowing what NOT to change is as valuable as knowing what to change.
came in looking for an exit upgrade. learned my current one is load-bearing.
3/3 — hobby project. not registered with SEC/CFTC/FINRA. nothing here is investment advice. #NFA
retro-tested 3 candidate exits — trailing chandelier, HTF momentum flip, naive +2/−1 fixed bracket (the control) — against my bot's 34 real prior entries using historical exchange candles.
the CONTROL lost money on the same entries where my live exit makes money.
2/3
you can't A/B test a trading bot's exit logic with live capital. every trade has one outcome, not three.
so i built a shadow framework. same entry signal, log-only mirrors of N candidate exits, ranked offline against live results.
1/3
Day 1 of 10x leverage on my Hyperliquid bot (a hobby project I've been chipping away at for 7 months).
First trade after deploy: SHORT INJ → flip to LONG → TP hit within $0.001 of target. Net +$1.69 on a $200 bag in 6h 51m.
Small money. Big confirmation that the new config works.
Config change shipped last night:
• Leverage 5x → 10x
• TP 10% → 5% (same $ PnL, half the time-in-trade)
• SL 5% → 2.5% (R:R unchanged at 2:1)
Math: shorter time-in-trade = less exposure to noise.
Risk: tighter stops = more chop kills, so a regime filter (ADX>25) gates entry to keep us out of choppy markets.
Sequence (UTC):
02:00:13 — Bot opens SHORT 14.3 INJ at $3.5058. ADX=27, regime=trending.
04:40:25 — Opposite signal fires on the same asset. The flip logic kicks in:
• Closes SHORT at $3.5671 (-$0.88)
• Waits for margin to settle (~8 seconds)
• Opens LONG 14.0 INJ at $3.5679
The flip mechanic is the part most retail bots get wrong. Hyperliquid's withdrawable balance lags the close by a few seconds — naive code that re-enters immediately will get rejected for insufficient margin every time.
06:51:33 — TP target was entry × 1.05 = $3.7463.
Trigger fired at $3.74735. Closed at $3.7516 (+$2.55).
Within $0.001 of the calculated target. No slippage on the limit. INJ is a fast-mover but limit-on-touch worked fine.
What I'm sharing isn't the dollars (it's $200 — testing).
It's the system behavior:
1) Flip logic worked end-to-end without margin starvation
2) TP hit exactly on the new config
3) Loss leg was small + contained
4) Regime filter let both directions through (ADX stayed >25 throughout)
Stack: TypeScript + Node 20 + Docker on a $24 droplet. HL SDK is @nktkas/hyperliquid (handles WebSocket reconnects cleanly).
A surprising amount of the work has been mundane infrastructure: the margin settlement wait, position reconciliation against the exchange, container health monitoring. The strategy code is the easy part.
Happy to answer questions about the flip mechanic, the regime filter, or
how I'm sizing positions on different leverage tiers.
---
Not financial advice. For entertainment and educational purposes only. I'm a US citizen sharing a personal hobby project — not registered with the SEC, CFTC, FINRA, or any other regulator. Trading perpetual futures
with leverage is high-risk and you can lose more than your deposit.
Don't copy these trades.
It was around now in 2020 that @ScrillaVentura told me to come to his house to FINALLY have a serious talk with me about Bitcoin Art
FULL DISCLOSURE, he told me a bunch of times before but i am an idiot
So FASTFORWARD to now... & the Bitcoin Sci Fi novel i wrote is at his house
@ChipotleJobs the crew handbook Chipotle Confidential number and https://t.co/81i7TLITJt is down / NOT working . @ChipotleTweets Can I please have someone from HR contact me? I need to discuss an HR concern ASAP
🚀 Just shipped the backend MVP for Aardmark — a PWA that turns voice notes into deliverables.
✅ Built with Vercel + Supabase + OpenAI
✅ Endpoints live: session creation, audio upload, deliverable generation
✅ Deliverables stored securely with signed URLs
🎯 Next: real transcription, multi-format outputs, and live AI replies.
⚡ Excited to keep building in public — if you’re working on AI-powered productivity tools, let’s connect and share lessons learned!
#AI #Supabase #Vercel #OpenAI #PWA #IndieHackers #BuildersInPublic #MVP
Someone talked me into joining the @DGBventures#FF league. Other than https://t.co/uJnogUv9Dj, Joe Montana, and Jerry Rice. I know nothing about Football, but I'm ready. LoL