Skip to content

README.md

Latest commit

 

History

History
118 lines (91 loc) · 6 KB

README.md

File metadata and controls

118 lines (91 loc) · 6 KB

Earnings Summarization

Overview

The Earnings Summarization use case analyzes earnings calls for capital markets professionals by coordinating transcript processing, financial metric extraction, and management sentiment analysis. It structures raw transcripts, extracts key financial metrics with guidance changes, and assesses management tone -- producing actionable investment insights for portfolio managers and analysts.

Business Value

  • Rapid analysis -- parallel processing compresses hours of manual earnings call review into minutes
  • Metric precision -- automated extraction of revenue, EPS, margins, growth rates, and guidance changes with consensus comparison
  • Sentiment objectivity -- five-level sentiment rating provides consistent, bias-free assessment of management tone
  • Investment-ready output -- structured response includes investment implications and notable management quotes for immediate use
  • Comprehensive coverage -- distinguishes between prepared remarks and Q&A, tracking sentiment shifts between sections

Architecture

graph TB
    Request["Client Request"] --> Runtime["AgentCore Runtime"]
    Runtime --> Orchestrator["Orchestrator"]
    Orchestrator --> Transcript["Transcript Processor<br/><small>Structure & organize earnings transcripts</small>"]
    Orchestrator --> Metric["Metric Extractor<br/><small>Financial metrics & guidance changes</small>"]
    Orchestrator --> Sentiment["Sentiment Analyst<br/><small>Management tone & confidence</small>"]
    Transcript --> Bedrock["Amazon Bedrock<br/>(Claude)"]
    Metric --> Bedrock
    Sentiment --> Bedrock
    Transcript --> S3["S3 Sample Data"]
    Metric --> S3
    Sentiment --> S3
    Transcript --> Synthesis["Result Synthesis"]
    Metric --> Synthesis
    Sentiment --> Synthesis
    Synthesis --> Response["Response"]
Loading

Directory Structure

use_cases/earnings_summarization/
├── README.md
└── src/
    └── strands/
        ├── __init__.py
        ├── config.py          # EarningsSummarizationSettings
        ├── models.py          # Pydantic request/response models
        ├── orchestrator.py    # EarningsSummarizationOrchestrator + run_earnings_summarization()
        └── agents/
            ├── __init__.py
            ├── transcript_processor.py
            ├── metric_extractor.py
            └── sentiment_analyst.py

Agentic Design

The orchestrator uses a parallel fan-out pattern. In full mode, all three agents execute concurrently via asyncio.gather. Individual modes (transcript_only, metrics_only, sentiment_only) invoke a single agent. The orchestrator synthesizes results through a structured prompt that produces JSON with sentiment rating, key metrics, guidance changes, notable quotes, risks, and investment implications.

Agents

Agent Role Data Used Output
Transcript Processor Structures raw earnings call transcripts into organized sections; identifies prepared remarks vs Q&A; extracts key discussion topics by business segment Entity profile via s3_retriever_tool Structured transcript with sections, speakers, key topics, segment breakdowns
Metric Extractor Extracts key financial metrics (revenue, EPS, margins, growth rates); identifies guidance changes and forward-looking statements; compares against consensus estimates Entity profile via s3_retriever_tool Structured metrics with categories, values, YoY changes, guidance updates, beats/misses
Sentiment Analyst Analyzes management tone and confidence levels; detects sentiment shifts between prepared remarks and Q&A; compares against prior quarter calls Entity profile via s3_retriever_tool Sentiment rating, confidence assessment, tone shifts, notable language patterns

Data and Tools

  • Tool: s3_retriever_tool -- retrieves earnings call entity profiles and transcript data from S3
  • S3 data prefix: samples/earnings_summarization/
  • Model: Claude Sonnet (via Amazon Bedrock), temperature 0.1, max 8192 tokens
  • Config thresholds: sentiment_confidence_threshold=0.80, metric_extraction_confidence=0.85, max_transcript_length=100000

Request / Response

Request -- SummarizationRequest:

Field Type Description
entity_id str Earnings call identifier (e.g., EARN001)
summarization_type SummarizationType full, transcript_only, metrics_only, sentiment_only
additional_context str | None Optional context

Response -- SummarizationResponse:

Field Type Description
entity_id str Entity identifier
summarization_id str Unique summarization UUID
timestamp datetime Summarization timestamp
earnings_overview EarningsOverview | None Sentiment rating, key metrics dict, guidance changes, notable quotes, risks identified
recommendations list[str] Investment implications
summary str Executive summary
raw_analysis dict Raw agent output

Quick Start

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

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

Sample Data

Located at data/samples/earnings_summarization/

Entity ID Company Ticker Quarter Description
EARN001 TechGrowth Inc TGRO Q4 2024 Technology sector, revenue $2.8B, EPS $1.45 (consensus $1.38), gross margin 62%, operating margin 28%, 18% YoY growth, prior quarter sentiment positive

Related Documentation