Spaces:
Sleeping
Sleeping
|
Download architecture.md from PROSTLE/StockSense-AI: direct link, hf CLI and curl.
- Browser
- Download file 1.65 kB
-
https://huggingface.co/spaces/PROSTLE/StockSense-AI/resolve/main/architecture.md
- Command line
-
hf download hf://spaces/PROSTLE/StockSense-AI/architecture.md
-
curl -L -o architecture.md https://huggingface.co/spaces/PROSTLE/StockSense-AI/resolve/main/architecture.md
1.65 kB
StockSense AI — System Flow
This flowchart describes how a single prediction request is processed through the system.
graph LR
%% User Entry
User((User)) -- "Enters Ticker" --> API[FastAPI Entry Point]
%% Data Layer
subgraph "Data Acquisition"
API --> YF[yfinance: Market OHLCV]
API --> News[News Aggregator: RSS/APIs]
end
%% Engine Layer
subgraph "AI Core Processing"
YF --> Indicators[Technical Signals: RSI, MACD, SMA]
YF --> LSTM[Deep Learning: LSTM Pattern Recognition]
News --> FinBERT[NLP: FinBERT Sentiment Scoring]
News --> Gemini[LLM: Gemini Market Reasoning]
end
%% Synthesis Layer
subgraph "Decision Engine"
Indicators & LSTM & FinBERT & Gemini --> Blender[5-Factor Weighted Blender]
Blender --> Regime[Crisis/Volatility Detection]
end
%% Output Layer
Regime --> Dashboard[[Final User Dashboard]]
Dashboard -- "JSON/Chart" --> User
Flow Components:
- User Request: Initiated when a ticker (e.g.,
RELIANCE.NS) is searched. - Data Acquisition: Simultaneously fetches 2 years of price history and current news.
- AI Logic:
- LSTM: Predicts price movement based on historical "waves."
- Technical Signals: Checks momentum and volume.
- FinBERT: Measures if the news is "Bullish" or "Bearish."
- Gemini: Provides a human-like explanation of why the score is what it is.
- Blender: Adjusts weights dynamically (e.g., news counts more if volatility is high).
- Dashboard: Returns the 5-day forecast, confidence level, and risk assessment.