SpeedRef — David’s Skills Reference

A highly scannable, pedantic technical reference for my engineering stack.

Welcome to SpeedRef. This is my personal knowledge base — built for fast retrieval, not for tutorials. It documents the exact architectures, patterns, and implementations I rely on across a production-grade Python + TypeScript stack (FastAPI · LangGraph · Postgres · React/Next.js).

Browse by Category

🚀 How to Use — The SpeedRef Spec

Every page in this reference adheres to a rigid, highly scannable layout to maximise signal density:

• WHAT (Concept Overview). A precise one-sentence definition of the technology, library, or pattern.
• Project Context. How I used this exact skill in fca-support-agent, the canonical multi-agent project these pages are distilled from.
• HOW (Quick Reference Blocks). Isolated chunks of code taken verbatim (or near-verbatim) from production. H3 per block, with bullet-pointed WHY beneath each.
• Common Pitfalls. 1–2 errors mid-level engineers hit and the exact fix.
• Real-World Interview Prep. Three current mid-to-senior interview questions with concise technical answers.

⭐ Featured Patterns

If you’re new here, start with the canonical backbone patterns:

• LangGraph Multi-Agent Streaming — Stream node-by-node tuples from a MessageWorkflow to the UI without losing routing context.
• LangGraph Checkpointing (AsyncPostgresSaver) — Make graphs survive process restarts and pause for human review; deadlock-safe DDL on first init.
• FastAPI SSE Streaming — Per-request ContextVar queue so the SSELogHandler can transparently pipe every log line into the live stream.
• SQLAlchemy Async Pooling — Environment-aware engine: NullPool for tests, pool_size + pool_pre_ping for prod.
• Multi-Stage Dockerfile — Slim base, CPU-only PyTorch install before requirements.txt, Spacy model baked into the image.
• Presidio PII + Lakera Jailbreak Defense — Two-layer outbound safety net: Spacy en_core_web_lg (Presidio) + Lakera Guard API.