ShreyaDesk AI

Research and execution need different boundaries. ShreyaDesk AI keeps its equity scanner read-only and routes eligible signals through a separate paper or live execution workflow. The dashboard brings together broker profile information, holdings, positions, orders, GTTs, stock research, and an analytics journal without treating a research signal as permission to trade.

The data layer intersects the official NSE equity list with Kite instruments and caches completed daily candles in SQLite. Same-day scans reuse that history, while validated quote overlays supply developing-session data without overwriting finalized candles. OCC and Supertrend scanning, daily RSI filters, and confirmed three-session signals share the underlying market data. Independent stock search provides charts, technical indicators, and transparent weighted research scores; missing inputs remain unavailable rather than being invented.

Execution is deliberately gated. Every backend startup leaves live trading paused. LIMIT orders require a verified broker account, a valid trading session, fresh signals, available allocation, and risk checks. Durable BUY reservations prevent overlapping entries and enforce daily submission limits. Cumulative loss monitoring and the kill switch persist independently of editable settings, while account mismatches pause execution instead of silently reassigning positions or orders.

A shared, versioned NSE session calendar coordinates cache preparation, scans, and risk-history coverage. Unknown years or unverified special-session hours block execution. The regular timetable prepares the cache at 09:00 IST, runs a results-only scan at 12:00, and permits the 15:00 scan to enter the existing execution checks. Telegram notifications run on a separate worker so network delays do not block the trading or safety path.

The application is intended for private local use or controlled deployment, with backend-only broker credentials and persistent storage. The supplied documentation includes backend unit tests, frontend builds, and Playwright browser checks. Its recorded warm-cache benchmark processed 2,542 symbols in 17.92 seconds with six quote calls and no historical-data calls; order dispatch was blocked and optional fundamentals enrichment was disabled. This is a documented test result, not a latency guarantee or evidence of trading returns. Paper-mode testing comes first; the project does not provide financial advice.