FiberFits AI Nutrition Platform

Problem and Context

• Nutrition analysis had to be structured and deterministic enough for product workflows, not only free-form AI text.
• The product needed near-real-time progress feedback while AI analysis ran asynchronously.

Built

• .NET 9 Clean Architecture API with Clerk JWT auth, EF Core, and config-as-data.
• Async OpenAI nutrition pipeline using channel queue, worker, JSON Schema outputs, and SSE updates.
• Next.js 15 app-router UI with HeroUI/Tailwind, typed API clients, next-intl, and resilient UX fallbacks.

Architecture

• MealEntry lifecycle and nutrition data modeled with domain invariants.
• Unit normalization, rowversion concurrency, and safe reprocessing rules.
• Frontend locale/timezone handling to avoid hydration mismatch and keep formatting correct.

Integrations

• OpenAI structured outputs through JSON Schema contracts.
• Clerk authentication, typed API clients, and next-intl localization.

Reliability, Security, and Operations

• Channel queue + worker pipeline with safe reprocessing rules and rowversion concurrency.
• SSE updates, optimistic fallbacks, and resilient image handling for unstable client/network states.

Operations

• Docker-based Linux deployments behind Cloudflare, Nginx, and Docker Compose.
• CI for lint, type-check, and build.
• Image compression, optimistic updates, localStorage fallbacks, and staged progress overlays.

Technology Stack