feat: Create Python SDK with auto-instrumentation

[haasonsaas]

🚀 Feature Request

Description

Create a Python version of eval2otel with auto-instrumentation capabilities, following OpenTelemetry's new instrumentation-genai patterns.

Motivation

Python is the dominant language in the AI/ML ecosystem. OpenTelemetry has started releasing Python instrumentation libraries for GenAI, and eval2otel should provide a Python SDK that aligns with these standards while offering zero-code instrumentation options.

Proposed Implementation

1. Core Python Package

   • Port TypeScript functionality to Python
   • Use Pydantic for type safety and validation
   • Follow OpenTelemetry Python conventions

2. Auto-Instrumentation

   • Support opentelemetry-instrument command
   • Automatic patching of popular AI libraries:
     • OpenAI
     • Anthropic
     • Google GenerativeAI
     • AWS Bedrock
     • Cohere
     • Hugging Face

3. Configuration

   from eval2otel import Eval2Otel, EvalResult
   
   # Initialize
   eval2otel = Eval2Otel(
       service_name='my-ai-service',
       capture_content=True,
       sample_content_rate=0.1
   )
   
   # Auto-instrument
   from eval2otel.auto import instrument_all
   instrument_all()
   
   # Or specific providers
   from eval2otel.instrumentations import instrument_openai
   instrument_openai()

4. Zero-Code Usage

   # Run with auto-instrumentation
   opentelemetry-instrument \
     --traces_exporter otlp \
     --metrics_exporter otlp \
     --service_name my-service \
     python main.py

5. Environment Variables

   • OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT
   • EVAL2OTEL_SAMPLE_RATE
   • EVAL2OTEL_REDACT_PII

Deliverables

• Python package on PyPI
• Auto-instrumentation support
• Documentation and examples
• Integration tests
• Type hints and Pydantic models

References

• OpenTelemetry Python GenAI Instrumentation
• OpenTelemetry Python Contrib

Priority

High - Python support is critical for AI/ML ecosystem adoption

[haasonsaas]

Implemented the Python SDK and auto-instrumentation surface in two merged PRs:

• Add RAGAS DeepEval and Python OTLP hooks #43 added optional Python OTLP span emission, content events, env-driven sampling/PII redaction, provider instrumentation handles, and Python docs.
• Add Python auto-instrument entry point #44 added the opentelemetry_instrumentor entry point for opentelemetry-instrument, optional Pydantic validation models, package metadata updates, and release/package smoke coverage.

The only remaining piece is the actual PyPI release/trusted-publishing setup, which is now tracked separately in #46 so this implementation issue can close cleanly.