Call Transcript Analysis

A FastAPI service that analyzes call center transcripts using the Anthropic Claude API. Submit a transcript and receive a structured QA score, call summary, and sentiment classification.

Features

• QA Scoring — 1–10 score based on call quality
• Summarization — 2–3 sentence summary of the call
• Sentiment Analysis — positive, neutral, or negative
• Structured JSON responses validated with Pydantic
• Retry logic with exponential backoff for transient API failures
• JSON structured logging with per-request IDs

Project Structure

app/
├── main.py                  # FastAPI entry point
├── api/
│   ├── routes.py            # POST /analyze-transcript endpoint
│   └── middleware.py        # Request logging middleware
├── services/
│   └── claude_service.py    # Claude API interaction
├── models/
│   └── schemas.py           # Request/response Pydantic models
└── core/
    ├── config.py             # Settings (env-based)
    ├── logging.py            # JSON logger
    └── resilience.py         # Retry decorator

Requirements

• Python 3.12+
• An Anthropic API key

Setup

1. Clone and create a virtual environment

git clone <repo-url>
cd call-transcript-analysis
python -m venv .venv
source .venv/bin/activate  # Windows: .venv\Scripts\activate

2. Install dependencies

pip install -r requirements.txt

3. Configure environment

cp .env.example .env

Edit .env:

ANTHROPIC_API_KEY=your-api-key-here
CLAUDE_MODEL=claude-sonnet-4-6
CLAUDE_TIMEOUT_SECONDS=30
CLAUDE_MAX_RETRIES=3

Running the Server

uvicorn app.main:app --reload

The API will be available at http://localhost:8000.

Interactive docs: http://localhost:8000/docs

API Reference

POST /analyze-transcript

Analyzes a call center transcript.

Request

{
  "transcript": "Agent: Thank you for calling. How can I help?\nCustomer: ..."
}

Field	Type	Constraints
transcript	string	1–50,000 characters, non-blank

Response 200 OK

{
  "qa_score": 8,
  "summary": "The agent resolved the customer's billing issue efficiently and with a friendly tone.",
  "sentiment": "positive"
}

Field	Type	Values
qa_score	integer	1–10
summary	string	1–2 sentence call summary
sentiment	string	"positive", "neutral", "negative"

Error responses

Status	Reason
422	Invalid or missing transcript input
503	Claude API unavailable or misconfigured

Resilience

Transient failures (rate limits, timeouts, server errors) are retried automatically with exponential backoff. The number of retries is controlled by CLAUDE_MAX_RETRIES (default: 3).

Tech Stack

Layer	Library
Web framework	FastAPI + Uvicorn
Validation	Pydantic v2
AI	Anthropic SDK (claude-sonnet-4-6)
HTTP client	HTTPX
Retry	Tenacity
Config	pydantic-settings + dotenv