Summary

• Add the AI Financial Assistant module with authenticated endpoints for:
  • Transaction categorization
  • Receipt OCR and structured data extraction
  • Natural-language financial Q&A
  • Spending trend and anomaly analysis
  • Budget and saving recommendations
• Introduce an extensible AIProvider interface and Gemini REST adapter.
• Support API key rotation and failover across multiple Gemini keys.
• Add dedicated AI rate limits, request timeouts, output token limits, and bounded financial context.
• Validate all HTTP inputs and AI structured outputs with Zod.
• Separate HTTP validation, AI response validation, and provider JSON schemas.
• Exclude profile data, receipt URLs, and transaction locations from AI context.
• Process OCR uploads in memory without persisting files.
• Add environment configuration examples, error codes, routes, and complete Swagger documentation.
• Update project documentation and architectural memory.

Verification

• pnpm.cmd run build
• Swagger specification serialization and AI endpoint verification
• git diff --check

Database Changes

No Prisma schema changes or database migrations are required.

Deployment Notes

• Configure GEMINI_API_KEYS as a comma-separated list of Gemini API keys.
• AI endpoints return HTTP 503 when the provider is not configured.
• The current AI rate limiter is process-local; multi-instance deployments should use a shared Redis-backed limiter or API gateway quota.