AI concept explainer / chat assistant with RAG

Summary

Chat UI trained on prompts + our 600+ posts as context (via RAG). "Why is my pod CrashLoopBackOff?" → returns an answer with citations back to our posts. Turns the content archive into an on-demand expert.

UX

• `/ask` page with a chat interface
• Each answer cites the 3-5 most relevant posts it drew from, linked inline
• Optional follow-up suggestions ("Related: how to debug imagePullBackOff")
• Answer quality feedback (thumbs up/down) so we can tune

Architecture

Static site constraint means we need an API layer:

• Cloudflare Workers + Workers AI (@cf/meta/llama-3.1-8b-instruct for drafting, @cf/baai/bge-base-en-v1.5 for embeddings)
• Embeddings for all posts/guides/exercises pre-computed at build time, stored in Vectorize or KV
• Chat endpoint: takes user query, embeds it, retrieves top-k, fills prompt template, streams response

Alternative: Anthropic API with citations. Higher quality, higher cost. Worth comparing.

Brand angle

• Positions DevOps Daily as "AI-native DevOps learning"
• Strong sponsor story ("Chat powered by X")
• Differentiator vs generic ChatGPT: answers grounded in our content only, so it's less prone to hallucination on DevOps-specific topics

Cost considerations

• Cloudflare Workers AI: ~free tier generous enough for early traffic
• If using Anthropic: need rate limits + caching layer
• Could require email (newsletter signup) to unlock the chat — turns chat into a lead gen tool

Phases

1. Phase 1 — pre-computed FAQ / concept explainer (no LLM at runtime; just semantic search → top matching post)
2. Phase 2 — full RAG chat with streaming
3. Phase 3 — conversational memory, code generation, scenario walkthroughs

Start by reviewing the existing static-site structure and content archive, then scope the proposed /ask page and chat endpoint. Clarify whether Phase 1 semantic search or the full RAG chat is the intended milestone; completion should specify the selected architecture, retrieval behavior, citations, and feedback or streaming requirements.