Finsight Nexus is an enterprise-grade, Bloomberg-style quantitative AI terminal. It bridges the gap between raw financial market data and actionable options trading strategies by leveraging a Zero-Hallucination RAG Pipeline and a LangGraph-powered Multi-Agent Execution Desk.

💡 Architectural Highlight: This system goes beyond standard LLM wrappers. It features autonomous agents that dynamically generate Python quantitative models, execute them in a local sandbox, and utilize the mathematical outputs to formulate mathematically-backed options playbooks.

• Semantic Signal Extraction: Ingests up to 48 hours of live macro and ticker-specific news firehose, utilizing vector similarity search to extract the Top 10 highest-signal headlines.
• Strict Grounding Guardrails: Prompt-engineered with absolute zero-knowledge-leakage constraints. The AI is forced to formulate thesis strictly on retrieved news and real-time technical indicators (RSI, MACD, SMA20).
• Evaluation Badge: Features a UI-integrated Faithfulness 10.0/10 health check badge to guarantee institutional-level trust.

A 4-node sequential state machine (Ledger) that mimics a real Wall Street quant desk:

1. Macro Strategist Agent: Formulates a baseline thesis utilizing VIX and price action.
2. Risk Quant Agent: Translates the thesis into executable Python code (e.g., implied daily volatility models: VIX / sqrt(252)).
3. Execution Sandbox (PythonREPL): Dynamically executes the AI-generated code to prove the mathematical viability of the thesis.
4. Head Trader Agent: Formulates the final options derivative strategy (e.g., Long Strangle, Iron Condor) heavily weighted by the exact output from the Sandbox.

• I/O Bottleneck Eradication: Integrates concurrent.futures.ThreadPoolExecutor to handle 30+ simultaneous Google Translate API requests, dropping translation latency from ~30 seconds to under 1.5 seconds.
• Dynamic Model Routing: Features a custom fail-safe radar (_get_dynamic_model) to automatically scan and hook into the active Gemini API endpoints, preventing 404 NOT_FOUND LangChain region lock errors.

• 100% elastic UI supporting seamless English (EN) and Chinese (CN) toggling.
• Not just static UI translation: System state, Multi-Agent thinking processes, and final RAG outputs are dynamically prompted to output strictly in the selected target language.

• Frontend / Dashboard: Streamlit, Plotly (Advanced Subplot Candlesticks, MACD, Volume)
• AI Orchestration: LangChain Core, LangGraph, Google Generative AI (Native SDK)
• Data Ingestion: yfinance, beautifulsoup4, requests
• Concurrency & i18n: concurrent.futures, deep-translator
• Execution Sandbox: langchain_experimental.utilities.PythonREPL