Slope: RAG-Powered Weight Coach on Managed Services

I built Slope, a privacy-first weight management app that leverages Retrieval Augmented Generation (RAG) to deliver personalized AI coaching without compromising user data.

Read the full blog post for a deep technical dive.

The core insight: effective weight coaching isn't about rigid meal plans—it's about consistent reflection, pattern recognition, and personalized guidance that adapts to your unique approach. Slope injects your personal context into every AI interaction, keeping all identifying data on your own servers.

I architected the entire system using managed services for cost efficiency (~$8/month): Vercel for Next.js hosting, Neon for serverless PostgreSQL, Upstash for Redis, and Render for background workers. The RAG implementation carefully constructs prompts with user context, recent history, weight data, and calorie tracking to generate genuinely helpful insights grounded in real data.

Key technical achievements: BullMQ job queue for async LLM operations, Next.js 15 server components for optimal performance, PostHog + Helicone for observability, and a custom design system built on Radix UI primitives.