Real-Time Candlestick Pattern Detector -- 2

I need a compact solution that plugs into my existing real-time market feed and continuously scans candlestick data for three specific formations: Head and Shoulders, Double Top/Bottom, and Cup and Handle. When any of these patterns complete (or are in a well-defined final stage), the system should raise an instant alert that I can route to a webhook, email, or simple on-screen notification.

The job centres on the recognition logic itself rather than feed acquisition; the streaming API is already available and delivers OHLCV ticks in JSON. A clean Python module using pandas, NumPy, or TA-Lib is preferred, but I am open to another language if it offers clear advantages and the code remains readable and easy to maintain.

Acceptance criteria:
• Processes live ticks with a maximum end-to-end delay of two seconds.
• Correctly identifies at least 90 % of test-set patterns provided after delivery.
• Emits a single, well-structured message per confirmed pattern, containing time stamp, symbol, pattern name, and confidence score.
• Includes concise documentation and a short demonstration script that runs against my sandbox feed.

Feel free to suggest enhancements such as adjustable sensitivity thresholds or a lightweight front-end chart overlay, but the core deliverable is the real-time pattern-recognition engine described above.