Intermediate5 lessons~125 min

Patterns in Production

The four architectures that cover most real systems

Speculative fan-out, confidence-gated routing, composite scoring, and intent routing — implemented properly, with the failure modes each one hides.

What you’ll be able to do

  • ✓Batch speculative questions without paying for them in latency
  • ✓Set risk-scaled confidence thresholds instead of one global number
  • ✓Move weighting logic out of prompts and into code you can tune
  • ✓Put a cheap classifier in front of an expensive LLM

Lessons

  1. 01

    Speculative Fan-Out

    free

    Ask everything you might need in one call. Discarding answers is cheaper than a second round trip.

    20m
  2. 02

    Confidence-Gated Routing

    paid

    The answer tells you what; confidence tells you whether to act. Tune a two-axis gate interactively.

    24m
  3. 03

    Composite Scoring

    paid

    Score dimensions independently, weight them in code. Change priorities without touching a prompt.

    22m
  4. 04

    Intent Routing

    paid

    Classify first, then route to deterministic code, a specialist LLM, or a human.

    20m
  5. 05

    Project: Moderation Pipeline

    paid

    An escalation ladder with per-action thresholds, an audit trail, and a human in the loop.

    40m