Skip to content

README.md

Latest commit

 

History

History
112 lines (85 loc) · 5.54 KB

README.md

File metadata and controls

112 lines (85 loc) · 5.54 KB

Email Triage

Overview

The Email Triage use case automates email processing for capital markets trading desks by coordinating email classification and action extraction. It categorizes incoming emails by type (client requests, trade instructions, compliance alerts, market updates, internal memos, meeting requests), assesses urgency and sender importance, extracts action items with deadlines, and produces prioritized triage summaries for traders and portfolio managers.

Business Value

  • Faster response times -- automated urgency assessment and action extraction ensure time-sensitive trade instructions are surfaced immediately
  • Prioritized workflow -- sender importance scoring and category classification help trading desks focus on high-impact communications first
  • Deadline tracking -- explicit extraction of deadlines and SLAs prevents missed time-sensitive requests
  • Reduced cognitive load -- structured triage summaries with recommended responses replace manual email scanning
  • Compliance awareness -- compliance alerts are automatically flagged and prioritized alongside trade-related communications

Architecture

graph TB
    Request["Client Request"] --> Runtime["AgentCore Runtime"]
    Runtime --> Orchestrator["Orchestrator"]
    Orchestrator --> Classifier["Email Classifier<br/><small>Category, urgency & topic identification</small>"]
    Orchestrator --> Extractor["Action Extractor<br/><small>Action items, deadlines & priorities</small>"]
    Classifier --> Bedrock["Amazon Bedrock<br/>(Claude)"]
    Extractor --> Bedrock
    Classifier --> S3["S3 Sample Data"]
    Extractor --> S3
    Classifier --> Synthesis["Result Synthesis"]
    Extractor --> Synthesis
    Synthesis --> Response["Response"]
Loading

Directory Structure

use_cases/email_triage/
├── README.md
└── src/
    └── strands/
        ├── __init__.py
        ├── config.py          # EmailTriageSettings
        ├── models.py          # Pydantic request/response models
        ├── orchestrator.py    # EmailTriageOrchestrator + run_email_triage()
        └── agents/
            ├── __init__.py
            ├── email_classifier.py
            └── action_extractor.py

Agentic Design

The orchestrator uses a parallel fan-out pattern with a two-agent design. In full mode, both agents execute concurrently via asyncio.gather. Individual modes (classification, action_extraction) invoke a single agent. The orchestrator synthesizes results through a structured prompt that produces JSON with category, urgency, sender importance, topics, actions, deadlines, and prioritization recommendations.

Agents

Agent Role Data Used Output
Email Classifier Classifies emails by content category (6 types), assesses sender importance (0-1 score) based on role/seniority/client relationship, determines urgency level, identifies key topics and workflow relevance Email profile via s3_retriever_tool Category, urgency level, sender importance score, key topics, workflow relevance
Action Extractor Extracts explicit and implicit action items, identifies deadlines and SLAs, captures key information (amounts, securities, counterparties, account numbers), prioritizes by urgency and business impact Email profile via s3_retriever_tool Action items with assignees, deadlines, key information extracted, priority ranking, suggested next steps

Data and Tools

  • Tool: s3_retriever_tool -- retrieves email profiles and content from S3
  • S3 data prefix: samples/email_triage/
  • Model: Claude Sonnet (via Amazon Bedrock), temperature 0.1, max 8192 tokens
  • Config thresholds: urgency_threshold=0.7, classification_confidence_threshold=0.8, max_actions_per_email=10

Request / Response

Request -- TriageRequest:

Field Type Description
entity_id str Email or batch identifier (e.g., EMAIL001)
triage_type TriageType full, classification, action_extraction
additional_context str | None Optional context

Response -- TriageResponse:

Field Type Description
entity_id str Email identifier
triage_id str Unique triage UUID
timestamp datetime Triage timestamp
classification ClassificationDetail | None Category, urgency, sender importance, topics, actions required, deadlines
recommendations list[str] Prioritization recommendations
summary str Executive summary
raw_analysis dict Raw agent output

Quick Start

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

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

Sample Data

Located at data/samples/email_triage/

Entity ID Sender Role Subject Description
EMAIL001 Portfolio Manager Urgent: Rebalance Request - Tech Sector Exposure Hedge fund PM requesting 15% tech sector reduction before market close, includes trade instructions attachment, priority indicators: urgent + market close deadline

Related Documentation