Skip to content

README.md

Latest commit

 

History

History
118 lines (91 loc) · 6.73 KB

README.md

File metadata and controls

118 lines (91 loc) · 6.73 KB

Trading Assistant

Overview

The Trading Assistant use case provides comprehensive trading intelligence for equity trading desks by coordinating market analysis, trade idea generation, and execution planning. It assesses market regime and conditions, generates trade ideas with risk-reward profiles and specific price levels, and plans optimal execution strategies including timing, venue selection, and order slicing -- producing actionable trading recommendations for desk traders.

Business Value

  • End-to-end trading workflow -- covers the full cycle from market analysis through idea generation to execution planning in a single request
  • Execution cost awareness -- execution planner estimates spread, impact, and timing costs with specific venue selection rationale
  • Risk-reward precision -- trade ideas include explicit entry, target, and stop levels with conviction scoring
  • Market regime classification -- market analyst provides objective regime assessment (bullish/bearish/neutral/volatile) grounded in data
  • Portfolio context -- trade ideas consider existing positions and correlation risk before recommending new exposure

Architecture

graph TB
    Request["Client Request"] --> Runtime["AgentCore Runtime"]
    Runtime --> Orchestrator["Trading Assistant<br/>Orchestrator"]
    Orchestrator --> Agent1["Market Analyst<br/><small>Analyzes price action, volume,<br/>liquidity, and market regime</small>"]
    Orchestrator --> Agent2["Trade Idea Generator<br/><small>Generates trade ideas with<br/>risk-reward profiles</small>"]
    Orchestrator --> Agent3["Execution Planner<br/><small>Plans execution strategy,<br/>timing, and venue selection</small>"]
    Agent1 --> Bedrock["Amazon Bedrock<br/>(Claude)"]
    Agent2 --> Bedrock
    Agent3 --> Bedrock
    Agent1 --> S3["S3 Sample Data<br/>(Market Data)"]
    Agent2 --> S3
    Agent3 --> S3
    Agent1 --> Synthesis["Result Synthesis"]
    Agent2 --> Synthesis
    Agent3 --> Synthesis
    Synthesis --> Response["Response"]
Loading

Directory Structure

use_cases/trading_assistant/
├── README.md
└── src/
    └── strands/
        ├── __init__.py
        ├── config.py          # TradingAssistantSettings
        ├── models.py          # Pydantic request/response models
        ├── orchestrator.py    # TradingAssistantOrchestrator + run_trading_assistant()
        └── agents/
            ├── __init__.py
            ├── market_analyst.py
            ├── trade_idea_generator.py
            └── execution_planner.py

Agentic Design

The orchestrator uses a parallel fan-out pattern with mode-dependent agent combinations. In full mode, all three agents execute concurrently via asyncio.gather. In trade_idea mode, the market analyst and trade idea generator run in parallel. In execution_plan mode, the market analyst and execution planner run in parallel. In market_analysis mode, only the market analyst runs. After agent execution, dedicated parser functions (parse_market_analysis) extract structured fields (market condition, urgency, confidence, key levels, trade ideas, execution notes) from LLM output into typed Pydantic models.

Agents

Agent Role Data Used Output
Market Analyst Analyzes real-time market data including price action, volume, and volatility; assesses order flow and institutional positioning; evaluates liquidity; identifies support/resistance levels and technical patterns; classifies market regime Trading profile and market data snapshot via s3_retriever_tool Market condition (BULLISH/BEARISH/NEUTRAL/VOLATILE), key price levels, volume/liquidity assessment, technical patterns
Trade Idea Generator Generates trade ideas based on market conditions and strategy parameters; evaluates risk-reward profiles with entry/target/stop levels; considers portfolio context and correlation risk; assesses conviction and sizing Trading profile and current positions via s3_retriever_tool Trade ideas with entry/target/stop, risk-reward ratio, conviction level, sizing recommendation, portfolio impact
Execution Planner Plans optimal execution considering market impact and order size; recommends timing based on liquidity patterns; suggests venue selection and order routing; designs slicing strategies (TWAP, VWAP, implementation shortfall) Trading profile and market conditions via s3_retriever_tool Execution strategy, timing/scheduling, venue selection, expected costs, risk mitigation

Data and Tools

  • Tool: s3_retriever_tool -- retrieves trading desk profiles, positions, market data snapshots, and risk limits from S3
  • S3 data prefix: samples/trading_assistant/
  • Model: Claude Sonnet (via Amazon Bedrock), temperature 0.1, max 8192 tokens
  • Config thresholds: market_impact_threshold=0.05, min_confidence_score=0.7, max_execution_slippage=0.02

Request / Response

Request -- TradingRequest:

Field Type Description
entity_id str Trading request identifier (e.g., TRADE001)
analysis_type AnalysisType full, market_analysis, trade_idea, execution_plan
additional_context str | None Optional context

Response -- TradingResponse:

Field Type Description
entity_id str Trading request identifier
analysis_id str Unique analysis UUID
timestamp datetime Analysis timestamp
market_analysis MarketAnalysisDetail | None Condition, urgency, confidence score (0-1), key levels, trade ideas, execution notes
recommendations list[str] Trading recommendations
summary str Executive summary
raw_analysis dict Raw agent output

Quick Start

# Deploy to AgentCore
USE_CASE_ID=trading_assistant ./scripts/deploy/full/deploy_agentcore.sh

# Test the deployment
./scripts/use_cases/trading_assistant/test/test_agentcore.sh

Sample Data

Located at data/samples/trading_assistant/

Entity ID Desk Strategy Description
TRADE001 Equity Trading Systematic Momentum Current positions in AAPL (5K shares, +$34K PnL), SPY (-2K shares short, +$5K PnL), NVDA (1.5K shares, +$38.7K PnL); SPX at 4783, VIX 13.2, 10Y yield 4.05%; risk limits: 5% max position, 20% max sector, $500K daily VaR, 10% max drawdown

Related Documentation