📄 PAPER TRADING — MONTH 1 OF 3

AI TRADING SYSTEM · NSE/BSE · AZURE

Trade-Lab.
AI That Trades
The Market.

An autonomous AI agent that trades NSE/BSE Nifty 50 stocks using four backtested swing strategies. Wakes at 9:00 AM IST every trading day, scans 32 stocks, consults Claude AI, and executes — fully without human involvement.

PYTHON CLAUDE AI DYNAMIC RISK AI UPSTOX API AZURE VM TELEGRAM

VIEW ARCHITECTURE → GITHUB (PRIVATE)

// SYSTEM STATUS

Phase Paper Trading

Market NSE/BSE Nifty 50

Stocks Scanned 32 Nifty 50

Capital ₹1,00,000

Jobs/Day 11 automated

AI Brain Claude (Anthropic)

Market Context Dynamic (4-State AI)

Data Feed Upstox API (SEBI)

Infrastructure Azure VM 24/7

System running since May 2026

SCROLL TO EXPLORE →

EMA Crossover ✦ Supertrend + MACD ✦ Dynamic Context ✦ Defensive Mode ✦ Breakout + Volume ✦ Momentum Pullback ✦ Claude AI ✦ Upstox API ✦ Azure VM ✦ NSE/BSE ✦ Risk Manager ✦ Self-Learning ✦ Swing Trading ✦ EMA Crossover ✦ Supertrend + MACD ✦ Dynamic Context ✦ Defensive Mode ✦ Breakout + Volume ✦ Momentum Pullback ✦ Claude AI ✦ Upstox API ✦ Azure VM ✦ NSE/BSE ✦ Risk Manager ✦ Self-Learning ✦ Swing Trading ✦

01 / The Problem

Human Limitations

Markets generate data every millisecond. Humans cannot scan 32 stocks simultaneously, enforce discipline under pressure, or wake at 9:00 AM IST without fail — every single day.

Bad Data = Bad Decisions

Yahoo Finance was an unofficial scraper — random 429 errors, 15–30 min stale prices, silent failures. Replaced with Upstox API: official SEBI-regulated, real-time, SLA-backed.

No Learning = No Edge

Static rule engines don't improve. Trade-Lab analyzes every closed trade across 10 dimensions and injects lessons directly into the AI's decision context — every week.

02 / Architecture

How It
Thinks.

AZURE VM 09:00 IST — Daily Trigger

HOLIDAY CHECK 16 NSE/BSE 2026 holidays

AI NIFTY CONTEXT Strong/Neutral/Mild/Bear

UPSTOX API 6mo OHLCV · Real-time

32 STOCKS Nifty 50 universe

30-MIN CACHE No redundant calls

EMA+RSI Score 1.0 · WR 63%

ST+MACD Score 1.2★ · WR 66%

BREAKOUT Score 0.8 · WR 57%

MOMENTUM Score 1.0 · WR 65%

SCORE GATE ≥ 2.0 ≥2 strategies must agree → else SKIP

EARNINGS GATE ±5 days → skip

SIGNAL FRESHNESS Stale >3d → skip

LEARNING MEMORY 10-dim lessons injected

CLAUDE AI decision · confidence · entry · SL · TP ~$0.005 per call

①

②

③

④

⑤

⑥

⑦

ALL 7 MUST PASS

PAPER BROKER +0.08% slippage

TRADE JOURNAL 30+ fields recorded

TELEGRAM Instant phone alert

03 / Daily Schedule

11 Jobs. Every Trading Day.

Fully autonomous — no human involvement required

09:00

Live Scan

Claude API

09:30

SL/TP Check

10:15

SL/TP Check

11:00

SL/TP Check

11:45

SL/TP Check

12:30

Midday Check

13:15

SL/TP Check

14:00

SL/TP Check

14:45

SL/TP Check

15:15

SL/TP Check

15:45

EOD Close

Sunday 20:00

Weekly Self-Learning Analysis

10-dim analysis · Updates context

04 / Strategies

Four Strategies.
One Consensus.

Score gate ≥ 2.0 required — at least 2 strategies must agree before Claude AI is consulted.

WEIGHT 1.0

EMA Crossover + RSI

0%

28 trades/yr · ~4 day hold

ENTRY: 9 > 21 > 50 > 200 EMA + RSI 40–70 + ADX > 20
SL/TP: entry − 1.5×ATR / entry + 3×ATR

WEIGHT 1.2 ★ BEST

Supertrend + MACD

0%

22 trades/yr · ~5 day hold

ENTRY: ST = Bullish + MACD > Signal line
SL/TP: ST line / entry + 2.5×ATR

WEIGHT 0.8

Breakout + Volume

0%

15 trades/yr · ~7 day hold

ENTRY: Close > 20d high + Volume > 1.3× avg
SL/TP: entry × 93% / entry × 115%

WEIGHT 1.0

Momentum Pullback

0%

20 trades/yr · ~4 day hold

ENTRY: Close > 20 > 50 EMA + RSI 45–62 + MACD hist > 0
SL/TP: entry − 1.5×ATR / entry + 3×ATR

EMA+RSI
1.0

+

ST+MACD
1.2

+

BKT
0.8

+

MOM
1.0

COMBINED SCORE ≥ 2.0

CLAUDE AI CALLED

05 / Self-Learning

The Bot That
Gets Smarter.

Every Sunday at 20:00 IST, Trade-Lab analyzes every closed trade across 10 dimensions and injects lessons directly into Claude's decision context.

TRADE CLOSESpaper_broker.sell() → journal

SUNDAY ANALYSISTradeAnalyzer runs 10-dim

CLAUDE CONTEXTLessons injected morningly

① Overall Performance

② By Strategy Combo

③ By Nifty Trend

④ By VIX Level

⑤ By Sector

⑥ By Individual Stock

⑦ By Day of Week

⑧ By Month

⑨ By RSI Zone

⑩ By AI Confidence

"In STRONG BULLISH: win rate is 72%. Confidence justified."

"Banking sector: only 44% WR — reduce exposure."

"RSI 50–62 entry zone: 71% WR vs 48% elsewhere."

06 / Risk Management

7 Rules. All Must Pass.

Hard-coded. Cannot be overridden. Not even by Claude.

✓

① BUY Signal Confirmed

Sanity check — decision must explicitly be BUY

✓

② Open Positions < 5

No overextension — max 5 simultaneous holdings

✓

③ Daily P&L > −3%

Day circuit breaker — stops all trades if down 3%

✓

④ Symbol Not Already Held

No averaging down — one position per symbol

✓

⑤ Sector Exposure < 40%

Diversification cap — no single sector dominance

✓

⑥ AI Confidence ≥ 65%

Raised from 55% — only high-conviction trades enter

✓

⑦ No Circuit Breaker Active

Hard block — system never trades through limit

// Dynamic Defensive Shield (AI Context)

Claude AI constantly evaluates the 10-day Nifty 50 trend, dynamically adjusting the bot's risk appetite and enforcing defensive filters to protect capital during downturns.

🟢 STRONG BULLISH / NEUTRAL

Normal operation. All 32 stocks scanned. Standard entry criteria apply.

🟡 MILDLY BEARISH

Restricted Mode Activated. Only scans defensive sectors (FMCG, Pharma, IT). Enforces strict oversold gates (RSI ≤ 40) & Volume > 20d avg.

🔴 STRONG BEARISH

Emergency Skip. Morning scan aborted immediately to protect capital in a crashing market.

⚙️ FAILSAFE

Deterministic rule-based fallback if Claude API goes down. The bot never runs blind.

// POSITION SIZING

risk_amount = capital * 0.02
shares = risk_amount / (entry - stop_loss)
cost = (shares * entry) + 20

// MARKET REALISM

Buy: entry price += 0.08% (slippage)
Sell: exit price -= 0.08% (slippage)
NSE Circuit: exit capped at prev_close * 0.90

07 / Tech Stack

Built
With.

Python

Core language — agent, strategies, risk, broker

Claude AI

Decision brain — entry, SL, TP, confidence scoring

Upstox API

SEBI-regulated NSE/BSE feed — OHLCV + live quotes

Azure VM

24/7 cloud deployment — 11 scheduled jobs/day IST

Telegram

Real-time phone alerts — BUY, SELL, daily report

GitHub Education

Free Azure credits (~$100) — ~9 months runtime

08 / Performance

Paper Results
So Far.

📄 PAPER TRADING IN PROGRESS — Month 1 of 3 · Live metrics will be published after 90-day validation

—

SHARPE RATIO

Publishing Month 3

—

MAX DRAWDOWN

Target: ≤ 15%

—

WIN RATE

Target: ≥ 55%

—

CLOSED TRADES

Target: ≥ 20

// LIVE TRADING CHECKLIST

90+ days of paper trading

Win rate ≥ 55%

Max drawdown ≤ 15%

≥ 20 closed trades

Start live with ₹50,000

Live trading begins Month 4 on Upstox · ₹50,000 initial

// API ECONOMICS

Market Condition	Calls/Day	Cost/Day	$4.77 credit lasts
Active bull market	8–12	$0.05	~3 months
Normal market	4–7	$0.025	~5 months
Quiet market	1–3	$0.01	16+ months

Note: Claude is only called when: trading day + score ≥ 2.0 + earnings clear + signal fresh

← BACK TO PROJECTS VIEW ON LINKEDIN →