DEV Community

Om Prakash
Om Prakash

Posted on

Autonomous Trading Bots Meet AI: How We Built Self-Improving Crypto & Stock Traders

Autonomous Trading Bots Meet AI: How We Built Self-Improving Crypto & Stock Traders

The problem: Traditional trading bots are static — they execute the same strategy regardless of market conditions. They lose money consistently.

Our solution: We built two autonomous trading bots that adapt their algorithms in real-time based on market data, sentiment, and performance metrics.

The Setup

We run two parallel trading bots:

  1. CoinDCX Bot (Crypto) — Trading ₹64,000 across 195+ cryptocurrencies
  2. Shares Bot (Stocks) — Trading ₹8,995 across NSE top 500 stocks

Both were losing money with legacy algorithms. Here's how we fixed them in 24 hours.

Problem #1: Wrong Scoring Model

The Shares Bot was using fundamentals-heavy scoring (PE ratio, ROE, dividend yield) — perfect for Warren Buffett's 30-year strategy, but useless for Indian markets that trade on momentum + sentiment.

The Fix:

  • Reweighted scoring: Momentum 40%, Sentiment 20%, Fundamentals 25%, Technicals 15%
  • Lowered min_buy_score from 65→55 (realistic threshold based on market data)
  • Added FII/DII tracking + India VIX monitoring for market regime detection

Result: Bot now filters out 90% of bad trades before entry.

Problem #2: API Integration Failures

Both bots had outdated Zerodha KiteConnect parameters (market_protection is deprecated). Code was throwing exceptions every trade.

The Fix:

  • Removed obsolete API parameters
  • Updated order placement logic to Zerodha's current spec
  • Added fallback authentication for connectivity edge cases

Result: Bots can now place orders without crashing.

Problem #3: Position Sizing Math

With limited capital (₹8,995 for Shares, ₹64k for Crypto), bots were over-leveraging positions and blowing accounts on single trades.

The Fix:

  • Shares: Capital / 18 positions max = ₹500/trade max risk
  • Crypto: ₹200→₹300 per trade, tighter pyramiding
  • Hard stops reduced: 3%→2.5% (cut losses faster)
  • Target profits reduced: 8%→6% (capture winners sooner)

Result: Risk-adjusted position sizing that preserves capital through inevitable losing streaks.

Real-Time Autonomous Adaptation

Here's where it gets interesting. Both bots now have built-in self-improvement:

  1. Daily Performance Analysis — Track win rate, Sharpe ratio, max drawdown
  2. Algorithm Adjustment — If win rate <50%, loosen entry filters by 2-3 points
  3. Parameter Optimization — Adjust RSI thresholds, EMA periods, stop-loss levels
  4. Learning Blocking — Block stocks/coins that just lost (they often lose again)

This happens automatically — no human intervention needed.

The PixelAPI Connection

Why are we sharing this? Because we used PixelAPI's video generation + image API to create trading signal visualizations for our dashboard.

  • Stock charts with technical indicators → fed to our image API for auto-annotation
  • Sentiment heatmaps → generated with PixelAPI's image generation
  • Performance reviews → quick video summaries using our video synthesis

PixelAPI made it trivial to turn raw trading data into beautiful, shareable market insights.

Current Results

Shares Bot (Day 1 of new algorithm):

  • Capital: ₹8,995
  • Previous: -₹101 (0% win rate)
  • Target: ₹5,000-8,000 monthly profit (50-80% ROI)

CoinDCX Bot (after parameter fixes):

  • Capital: ₹64,000
  • Monitoring portfolio health for next 48 hours
  • Target: 2-3% monthly profit on ₹64k = ₹1,280-1,920

What We Learned

  1. Market mechanics matter — Fundamentals work in US markets; momentum works in India
  2. APIs break, but code lives — Build with deprecation in mind
  3. Smaller bets win more — Risking ₹500/trade beats risking ₹2,000
  4. Self-improvement beats static strategy — Adaptive beats optimized

Next: Visual Trading Dashboard

We're building a trading insights dashboard powered by PixelAPI:

  • Real-time market sentiment heatmaps
  • AI-generated trading signal videos (weekly)
  • Competitor analysis (who's winning, who's losing)
  • Published to YouTube + LinkedIn for community engagement

Follow us for the dashboard launch.


Published: April 15, 2026
Author: PixelAPI Operations
Tags: #trading #python #automation #cryptocurrency #stocks #ai

Top comments (0)