AI Architecture

Architecture Blueprint

FlowDeck uses two complementary AI systems: the TradingAgents Graph for deep stock analysis and the Briefing Agent for personalized daily portfolio briefs.

Briefing Agent

User Daily Brief Pipeline

The Briefing Agent generates personalized daily (or weekly/custom) portfolio briefs by combining algorithmic context building with specialized LLM agents. It analyzes your subscribed stocks, identifies priority tickers based on attention scores, and produces a concise narrative with market context and actionable insights.

Pipeline Stages

Context Builder

Algorithmic

Algorithmic stage that loads portfolio, fetches market data, ranks tickers by attention score, and builds comprehensive context.

Inputs

•user_id
•digest_date
•portfolio_tickers
•span_type (daily/weekly/custom)

Outputs

•DigestContext with quotes, returns, news, fundamentals, platform reports, market movers, sector/peer data

Focus Selector Agent

LLM Agent

LLM agent that determines which tickers to focus on based on attention scores, user preferences, and portfolio context.

Inputs

•portfolio_tickers
•attention_scores
•user_note
•user_focus_tickers

Outputs

•focus_tickers (ordered list of priority tickers)

Ticker Interpreter Agent

LLM Agent

LLM agent that analyzes each priority ticker to explain what happened, identify the main driver, and compare to platform thesis.

Inputs

•ticker
•quote
•returns
•news
•fundamentals
•platform_reports
•sector/peer context

Outputs

•TickerInterpretation (explanation, driver classification, thesis comparison)

Market Interpreter Agent

LLM Agent

LLM agent that synthesizes market-wide context and explains relevance to the portfolio.

Inputs

•market_movers
•global_news
•ticker_interpretations
•portfolio_tickers

Outputs

•MarketInterpretation (market summary, portfolio relevance)

Narrative Writer Agent

LLM Agent

LLM agent that composes the final brief narrative with market highlights, key signals, what to watch, and risks/opportunities.

Inputs

•ticker_interpretations
•market_interpretation
•user_note
•narrative_style
•resources

Outputs

•narrative (structured or basic format)
•what_to_watch
•references

Key Features

✓Attention Scoring: Ranks tickers by absolute returns, abnormal moves, and recent news
✓Multi-Span Support: Daily, weekly, or custom date ranges
✓User Preferences: Optional user notes, focus tickers, and narrative style
✓Platform Integration: References FlowDeck reports and provides shareable URLs
✓Structured References: Tracks and cites news articles, feeds, and web sources

TradingAgents Graph

Deep Stock Analysis Pipeline

The TradingAgents Graph performs comprehensive stock analysis through a multi-phase debate system. Specialized analysts gather evidence, bull/bear researchers debate investment merits, and risk analysts evaluate tradeoffs before producing final recommendations.

Runtime default analyst selection: market → news → fundamentals → technical → sec (US only). Social analyst is available but not selected by default.

Schematic Workflow

Graph order from `GraphSetup.setup_graph` and `ConditionalLogic`.

Phase 1: Analyst Pipeline

Input State

ticker + trade_date + empty reports

→

Market Analyst

score + report

→

News & Sentiment Analyst

score + report

→

Fundamentals Analyst

score + report

→

Technical Analyst

score + report

→

SEC Analyst

US only

Each analyst loops with its `tools_*` node until no more tool calls, then graph clears messages and moves forward.

Phase 2: Investment Debate + Plan

Bull Researcher

writes bull argument

→

Bear Researcher

writes bear argument

→

Neutral Researcher

writes balanced argument

→

Debate Loop

3 * max_debate_rounds

→

Research Manager

investment plan + BUY/SELL/HOLD + score + returns

→

Trader

trader plan + TPS

Agent-by-Agent Contracts

Exact state contracts for each analyst, researcher, manager, and trader.

Phase 1: Analyst Pipeline

Specialized analysts gather market, news & sentiment, fundamentals, technical, and SEC/regulatory evidence.

Market Analyst

Analyst · default: Yes

Consumes

trade_date, company_of_interest, messages

Tools

get_ticker_data, get_ticker_quote, get_indicators

Emits

messages, market_report, market_score

Output shape

{
  "messages": [
    {
      "type": "AIMessage",
      "content": "<market_report>"
    }
  ],
  "market_report": "<string>",
  "market_score": "<int 1-5 | null>"
}

News & Sentiment Analyst

Analyst · default: Yes

Consumes

trade_date, company_of_interest, messages

Tools

get_ticker_quote, get_events, get_news, get_global_news, get_insider_transactions, get_reddit_company_social, get_polymarket_sentiment

Emits

messages, sentiment_report, sentiment_score

Output shape

{
  "messages": [
    {
      "type": "AIMessage",
      "content": "<sentiment_report>"
    }
  ],
  "sentiment_report": "<string>",
  "sentiment_score": "<int 1-5 | null>"
}

Fundamentals Analyst

Analyst · default: Yes

Consumes

trade_date, company_of_interest, messages

Tools

get_fundamentals, get_balance_sheet, get_cashflow, get_income_statement

Emits

messages, fundamentals_report, fundamentals_score

Output shape

{
  "messages": [
    {
      "type": "AIMessage",
      "content": "<fundamentals_report>"
    }
  ],
  "fundamentals_report": "<string>",
  "fundamentals_score": "<int 1-5 | null>"
}

Technical Analyst

Analyst · default: Yes

Consumes

trade_date, company_of_interest, messages

Tools

get_ticker_data, get_ticker_quote, get_indicators, detect_divergence, detect_regime, detect_support_resistance

Emits

messages, technical_report, technical_score

Output shape

{
  "messages": [
    {
      "type": "AIMessage",
      "content": "<technical_report>"
    }
  ],
  "technical_report": "<string>",
  "technical_score": "<int 1-5 | null>"
}

SEC Analyst

Analyst · default: Yes (US companies)

Consumes

trade_date, company_of_interest, messages

Tools

get_edgar_filing_content

Emits

messages, sec_report, sec_score

Output shape

{
  "messages": [
    {
      "type": "AIMessage",
      "content": "<sec_report>"
    }
  ],
  "sec_report": "<string>",
  "sec_score": "<int 1-5 | null>"
}

Phase 2: Bull/Bear/Neutral Debate + Plan

Bull, bear, and neutral researchers debate in round-robin, then the research manager weighs all three sides and emits the investment plan with the authoritative BUY/SELL/HOLD recommendation. The trader then turns it into an executable plan (TPS).

Bull Researcher

Researcher · default: Always

Consumes

market_report, sentiment_report, fundamentals_report, technical_report, investment_debate_state

Emits

investment_debate_state

Output shape

{
  "investment_debate_state": {
    "history": "<string>",
    "bull_history": "<string>",
    "bear_history": "<string>",
    "neutral_history": "<string>",
    "latest_speaker": "Bull",
    "current_response": "Bull Analyst: <string>",
    "count": "<int>"
  }
}

Bear Researcher

Researcher · default: Always

Consumes

market_report, sentiment_report, fundamentals_report, technical_report, investment_debate_state

Emits

investment_debate_state

Output shape

{
  "investment_debate_state": {
    "history": "<string>",
    "bull_history": "<string>",
    "bear_history": "<string>",
    "neutral_history": "<string>",
    "latest_speaker": "Bear",
    "current_response": "Bear Analyst: <string>",
    "count": "<int>"
  }
}

Neutral Researcher

Researcher · default: Always

Consumes

market_report, sentiment_report, fundamentals_report, technical_report, investment_debate_state

Emits

investment_debate_state

Output shape

{
  "investment_debate_state": {
    "history": "<string>",
    "bull_history": "<string>",
    "bear_history": "<string>",
    "neutral_history": "<string>",
    "latest_speaker": "Neutral",
    "current_response": "Neutral Analyst: <string>",
    "count": "<int>"
  }
}

Research Manager

Manager · default: Always

Consumes

investment_debate_state, market_report, sentiment_report, fundamentals_report, sec_report, technical_report

Emits

investment_debate_state, investment_plan, recommendation, recommendation_score, bull_summary, bear_summary, neutral_summary, expected_return_pct, bear_case_return_pct, bull_case_return_pct

Output shape

{
  "investment_debate_state": {
    "judge_decision": "<string>",
    "history": "<string>",
    "bear_history": "<string>",
    "bull_history": "<string>",
    "neutral_history": "<string>",
    "latest_speaker": "<string>",
    "current_response": "<string>",
    "count": "<int>"
  },
  "investment_plan": "<string>",
  "recommendation": "BUY | SELL | HOLD",
  "recommendation_score": "<int 1-5 | null>",
  "bull_summary": [
    "<string>"
  ],
  "bear_summary": [
    "<string>"
  ],
  "neutral_summary": [
    "<string>"
  ],
  "expected_return_pct": "<number | null>",
  "bear_case_return_pct": "<number | null>",
  "bull_case_return_pct": "<number | null>"
}

Trader

Trader · default: Always

Consumes

company_of_interest, investment_plan, market_report, sentiment_report, fundamentals_report

Emits

messages, trader_investment_plan, trader_recommendation, trader_tps_plan, trader_key_takeaways, sender

Output shape

{
  "messages": [
    {
      "type": "AIMessage",
      "content": "<trader_investment_plan>"
    }
  ],
  "trader_investment_plan": "<string>",
  "trader_recommendation": "BUY | SELL | HOLD | null",
  "trader_tps_plan": "<TPS JSON string>",
  "trader_key_takeaways": [
    "<string>"
  ],
  "sender": "Trader"
}

Final Logged Run Output

Final output structure persisted by `TradingAgentsGraph._log_state(...)`.

{
  "company_of_interest": "<string>",
  "trade_date": "<YYYY-MM-DD>",
  "market_report": "<string>",
  "market_score": "<int | null>",
  "sentiment_report": "<string>",
  "sentiment_score": "<int | null>",
  "fundamentals_report": "<string>",
  "fundamentals_score": "<int | null>",
  "sec_report": "<string>",
  "sec_score": "<int | null>",
  "technical_report": "<string>",
  "technical_score": "<int | null>",
  "investment_debate_state": {
    "bull_history": "<string>",
    "bear_history": "<string>",
    "neutral_history": "<string>",
    "history": "<string>",
    "current_response": "<string>",
    "judge_decision": "<string>"
  },
  "trader_investment_decision": "<string>",
  "trader_tps_plan": "<TPS JSON string>",
  "investment_plan": "<string>",
  "recommendation": "BUY | SELL | HOLD",
  "recommendation_score": "<int | null>",
  "expected_return_pct": "<number | null>",
  "bear_case_return_pct": "<number | null>",
  "bull_case_return_pct": "<number | null>"
}