Optimizing Guest-to-Login Data Transfer: From 4s to 100ms
How reordering async operations, splitting API calls, and deferring storage moves eliminated a 4-second login delay and fixed missing chat history.
Learnings
Hands-on records of learning AI tools and technologies
Python asynccontextmanager & FastAPI Lifespan Pattern
How @asynccontextmanager and yield divide startup/shutdown in FastAPI lifespan, and how dunder methods __aenter__/__aexit__ work as Python's context manager protocol.
LLM Fundamentals: Parameters, Embeddings, and Attention
How LLM parameters encode meaning, what embedding dimensions actually represent, and why Transformer attention is computed in parallel.
Generative AI Design Patterns Landscape and AI Engineer Career Positioning
A structured map of GenAI application patterns and a practical framework for deciding how deep to go — calibrated to AI Product Engineer vs AI Engineer roles.
AI Pipeline Patterns from LinguaRAG: RAG, Streaming, and Prompt Architecture
A comprehensive breakdown of AI pipeline concepts learned building LinguaRAG — a Korean-German textbook AI tutor using RAG, SSE streaming, multi-layer prompts, and pgvector.
PDF RAG Indexing: Unit Detection and Chunk Noise Filtering
How to reliably detect structured unit boundaries in a bilingual PDF and prevent boilerplate text from polluting RAG vector chunks.
PDF Indexing Pipeline: Unit Detection Guards and Copyright Filtering
Hard-won lessons from building a robust PDF chunker for a Korean-German textbook: multiple detection guards, line-level copyright stripping, and RAG behavior verification.
Supabase Auth + FastAPI JWT Verification: HS256 → ES256 Migration
New Supabase projects sign JWTs using the ES256 algorithm. How to verify with PyJWT + JWKS client and how to resolve macOS Python SSL issues.
RAG Architecture Fundamentals — pgvector, FastAPI, SSE Streaming, and Embedding Models
Core RAG concepts understood while planning LinguaRAG: offline/online phase separation, SSE streaming mechanics, prompt assembly, and the role of pgvector.
AI Engineering Fundamentals - Essential Concepts Before Learning AI Agents
Every concept in this document is explained through a single real-world app -- **"SupportBot" (AI-powered customer support system)**. Each section...
Spec Interview Plugin — AI-Driven Requirements Gathering with AskUserQuestion
A new Claude Code plugin that collects requirements by having **the AI interview the user**.
AI PE Learning Agent Project Setup
Building AI agents to learn AI engineering - the tool creation process itself becomes the learning journey. Instead of following tutorials, you solve...
Understanding Claude Code Marketplace and Plugins
Claude Code's marketplace is a **decentralized system**, not a centralized App Store.