WULF$12.44▲ +4.82%
NVDA$875.20▲ +3.41%
PLTR$24.88▲ +2.94%
MSTR$1,284.00▲ +2.17%
MU$128.33▲ +1.88%
HOOD$18.92▲ +1.54%
AAPL$189.42▲ +1.24%
TSLA$248.17▼ -0.83%
AMD$168.55▲ +2.04%
SPY$521.88▲ +0.62%
CIFR$8.14▲ +3.22%
BTDR$11.77▲ +5.10%
WULF$12.44▲ +4.82%
NVDA$875.20▲ +3.41%
PLTR$24.88▲ +2.94%
MSTR$1,284.00▲ +2.17%
MU$128.33▲ +1.88%
HOOD$18.92▲ +1.54%
AAPL$189.42▲ +1.24%
TSLA$248.17▼ -0.83%
AMD$168.55▲ +2.04%
SPY$521.88▲ +0.62%
CIFR$8.14▲ +3.22%
BTDR$11.77▲ +5.10%
VERSION 5 — FULLY DEPLOYED
// CIRCUIT & CO — PROJECT 002 — AI ALGORITHMIC TRADING SYSTEM

ALPHA
BOT

AI Algorithmic Trading System · v5

A fully autonomous AI trading system scanning 10,000+ US equities through a 24-signal pipeline — gradient-boosted ML, FinBERT sentiment, reinforcement learning, Federal Reserve language analysis, Level 2 order book, and more. Backtested at +290% total return with a Sharpe Ratio of 2.73. Running 24/7 on dedicated hardware with a real-time dashboard.

● LIVE 24/7 XGBoost · FinBERT ML · Q-Learning Sharpe 2.73 24 Signal Sources 19 Risk Gates
Running Live 24/7 Explore the System
Total Return (2yr Backtest)
+290%
$100K → $390K simulated
Alpha vs SPY
+249%
SPY returned +41% same period
Sharpe Ratio
2.73
Risk-adjusted return — institutional grade
Max Drawdown
-6.83%
Controlled downside across 2 years
Profit Factor
1.58
Win Rate 49.4% · 48 curated tickers
System Overview

By the Numbers

Analyses Logged
44,698+
Historical rows XGBoost trained on — plus 16,400+ live analyses since deployment
Equities Scanned
10,000+
Full US equity universe scanned every 5 minutes during market hours
Signal Sources
24
Weighted composite score summing to 1.00 — walk-forward optimized weekly
Risk Gates
19
Every gate must clear before a single order is placed — no exceptions
Signal Pipeline

24-Signal Architecture

Every ticker is scored across 24 independent signal sources. The weighted composite determines trade candidacy — then 19 risk gates make the final call.

01
XGBoost ML Agent
Gradient boosting model trained on 44,698 historical analysis rows — highest win-rate impact of any single agent
ML
02
FinBERT Sentiment Agent
ProsusAI FinBERT transformer model running locally — financial NLP sentiment on real-time news, far beyond keyword matching
TRANSFORMER
03
Reinforcement Learning Position Sizing
Q-learning agent that optimizes position size based on signal confidence and current portfolio state — adapts over time
RL
04
Multi-Timeframe Analysis
Parallel analysis across 5-minute and 1-hour Alpaca bars — confirms signals across timeframes before acting
QUANT
05
Level 2 Order Book Agent
Real-time bid/ask spread, order imbalance, and depth analysis via Alpaca — detects institutional flow before price moves
MICROSTRUCTURE
06
Federal Reserve Language Analysis
Scrapes and NLP-parses actual FOMC statements from federalreserve.gov — detects hawkish/dovish shifts in Fed language
NLP
07
Earnings Transcript NLP
SEC EDGAR earnings call transcript analysis — sentiment and keyword extraction around reporting events
NLP
08
WSB Crowd Sentiment Agent
PRAW Reddit integration monitoring WallStreetBets — captures retail momentum signals before they hit price action
SOCIAL
09
Cross-Asset Macro Agent
VIX fear index, gold, bond yields, and dollar index — reads macro risk environment and adjusts signal bias accordingly
MACRO
10
Market Regime Detection
Classifies current market as bull/bear/sideways/high-volatility — dynamically adjusts position sizing and signal weights
REGIME
11
On-Chain Crypto Data Agent
Fear & Greed index + CoinGecko data — crypto market sentiment as a leading indicator for risk-on/risk-off equity behavior
ON-CHAIN
12
Statistical Arbitrage Agent
Sector basket correlation analysis — identifies mean-reversion opportunities across correlated equity pairs
STAT ARB
13
Pairs Trading Agent
Monitors cointegrated stock pairs for divergence — generates signals when historical spread relationships break down
PAIRS
14
Self-Optimizing Weight System
Walk-forward weight optimization runs weekly — signal weights in weights.json automatically adjust based on recent performance
ADAPTIVE
15–24
+ 10 Additional Signal Sources
Technical analysis, fundamental scoring, portfolio VaR, supply chain correlations, market microstructure, knowledge graph ripple effects, and more — combined score sums to exactly 1.00
MULTI-FACTOR
Live Output

Bot in Action

alpha-bot v5 — raspberry-pi-5 — alphabot.service — ACTIVE
[INIT ] AlphaBot v5 — 24-signal pipeline — 10,247 equities loaded [MODEL ] XGBoost loaded — trained on 44,698 rows — accuracy: 67.3% [MODEL ] FinBERT loaded from local snapshot — ProsusAI/finbert [REGIME] Market regime: BULL — risk-on bias active [MACRO ] VIX: 14.2 (low fear) · Gold: neutral · Dollar: slight weakness [FED ] FOMC statement parsed — tone: neutral-dovish → +0.12 bias ──────────────────────────────────────────────────────────── [SCAN ] Cycle 3,847 — WULF flagged for deep analysis [XGB ] XGBoost prediction: 71.2% bullish probability → +0.82 [FINBRT] FinBERT news sentiment: +0.74 (8 articles analyzed) → +0.74 [L2 ] Order book: bid pressure 68% · spread: 0.12% · imbalance: bullish → +0.68 [WSB ] Reddit mentions: 1,204 (+480% 24h) · sentiment: bullish → +0.91 [SCORE ] WULF composite score (24 signals): 0.887 / 1.00 — STRONG BUY ✓ ──────────────────────────────────────────────────────────── [RISK ] Running 19 risk gates... [RISK ] All 19 gates cleared ✓ — RL position size: 3.8% portfolio [ROUTE ] Smart order routing: limit order @ $12.38 (below ask) [EXEC ] BUY WULF — 248 shares @ $12.38 — $3,070.24 — TP: $13.44 SL: $11.80 [LOG ] Trade #16,401 → SQLite · dashboard updated · weights.json scheduled [WATCH ] Monitoring 6 open positions · next scan in 4m 32s...
Architecture

Tech Stack

AI & Machine Learning
XGBoost + FinBERT + RL
XGBoost trained on 44,698 historical rows. FinBERT transformer running locally from ProsusAI snapshot. Q-learning reinforcement agent for position sizing. scikit-learn for additional classification models.
XGBoostFinBERTTensorFlowscikit-learnQ-Learning
Core & Deployment
FastAPI + Raspberry Pi 5 + systemd
FastAPI backend serving all signal agents via async endpoints. APScheduler runs the 5-minute scan cycle. Deployed on a Raspberry Pi 5 via systemd services — alphabot and alphabot-runner — running 24/7.
PythonFastAPIAPScheduleruvicornsystemdRaspberry Pi 5
Data & Frontend
Alpaca + Next.js Dashboard
Alpaca Markets API for paper trade execution and real market data. yfinance for historical feeds. PRAW for Reddit. SQLite logs all trades and analyses. Next.js live dashboard with real-time P&L, live mode toggle, and 30-second auto-refresh.
Alpaca APIyfinancePRAWSQLiteNext.js
Risk Management

19 Risk Gates

Every trade must clear all 19 gates. One failure kills the trade. No exceptions.

✓
Minimum Score Threshold
Composite score must exceed 75/100
✓
Portfolio Exposure Limit
Max position size enforced by RL agent
✓
Daily Loss Limit
Auto-halts if daily drawdown exceeded
✓
Liquidity Check
Minimum average daily volume required
✓
Market Hours Validation
No trades outside regular session
✓
Volatility Gate
Extreme volatility events block execution
✓
Duplicate Position Block
Cannot add to an existing open position
✓
Bid/Ask Spread Check
Spread must be within acceptable range
✓
Regime Compatibility
Trade direction must match market regime
✓
Agent Consensus Gate
Minimum signal agreement required
✓
News Blackout Check
Blocks trades around major breaking news
✓
Earnings Window Block
No new positions before earnings releases
✓
Sector Concentration
Prevents overexposure to one sector
✓
Market Cap Floor
Filters out penny stocks and illiquid names
✓
Stop Loss Validation
Valid SL level must be calculable pre-trade
✓
Reward:Risk Ratio
Minimum R:R enforced before every order
✓
EOD Force-Close
All positions closed before market end
✓
Capital Availability
Buying power confirmed before order placement
✓
Signal Freshness
Stale signals older than threshold rejected

Live Dashboard

The bot is running right now, 24/7, on dedicated hardware. The Next.js frontend shows real-time P&L, open positions, recent trades, and signal scores — pulling live data from the FastAPI backend every 30 seconds.

Live on Private Infrastructure
ALPHA BOT v5 — LIVE
────────────────────────────
Portfolio Value: $103,847.22
Today's P&L: +$1,204.88 (+1.17%)
Open Positions: 6
Trades Today: 14
Win Rate (30d): 52.3%
Next Scan: 2m 14s
────────────────────────────
Status: ● ALL SYSTEMS OPERATIONAL

See More Work

View the full Circuit & Co portfolio or get in touch directly.

← Back to Portfolio