Smart Inspector

# Case Study: App 01 — Smart Inspector ## 1. Overview Smart Inspector is a multi-modal vision inspection application. It allows users to upload images and automatically extracts structured captions, keyword tags, detected physical objects, and verbatim OCR text transcriptions. ## 2. Learning & Technical Goals - Integrate Azure OpenAI Responses API for multi-modal image analysis. - Enforce strict server-side image validation (magic bytes, dimensions, EXIF stripping). - Build a dual-layer cost protection model (rate limiting, concurrency caps, kill switch). ## 3. Architecture & Tech Stack - **Frontend:** Next.js 16, Tailwind CSS, TypeScript. - **Backend:** FastAPI, Python 3.12, Pillow, Azure OpenAI SDK. ## 4. Key Security & Cost Control Decisions - **Image Validation:** Uploads are checked via Pillow (`Image.open`, `verify`). Dimension bounds are enforced (max 8000x8000 px / 24M pixels). - **EXIF Stripping:** EXIF and camera metadata are stripped in memory prior to forwarding base64 payloads to Azure. - **Rate Limiting:** `InMemoryRateLimiter` enforces 3 req/min per IP and 100 global daily requests. - **Concurrency Control:** `asyncio.Semaphore` caps concurrent active analysis operations at 2. ## 5. Outcome Smart Inspector is fully hardened, responsive, internationalized in 3 languages (`pt-BR`, `en`, `de`), and safe for public demo deployment.