AI Processing Service

Worker service that consumes SQS messages, analyzes architecture diagrams with Gemini 2.0 Flash Vision, and sends structured reports to the report service.

Responsibilities

• Consume messages from SQS queue (diagram-analysis-queue)
• Download diagram files from S3
• Convert PDFs to images (pdf2image + poppler)
• Analyze diagrams with Google Gemini 2.0 Flash Vision (LangChain)
• Validate LLM output with guardrails (Pydantic schema + sanitization)
• POST results to report-service
• Update upload status in upload-service

AI Pipeline

SQS Message → Download S3 → (PDF→PNG) → Base64 → Gemini 2.0 Flash Vision → Guardrails → Report Service

Guardrails

• JSON schema validation (Pydantic)
• Severity/Priority enum enforcement
• String sanitization (trim, length cap)
• Retry on failure (2x)

Environment Variables

Variable	Description
GOOGLE_API_KEY	Google Gemini API key — obtenha em https://aistudio.google.com (required)
AWS_REGION	AWS region
S3_BUCKET_NAME	S3 bucket containing diagrams
SQS_QUEUE_URL	SQS queue URL to consume from
UPLOAD_SERVICE_URL	Internal URL of upload-service
REPORT_SERVICE_URL	Internal URL of report-service
INTERNAL_SERVICE_TOKEN	Token for service-to-service calls

Running locally

cp .env.example .env
# Edit .env — GOOGLE_API_KEY is required

pip install -r requirements.txt
python -m src.worker

Running tests

pytest tests/ -v --cov=src