Most confusion in AI engineering comes not from model math but from vocabulary — different teams mean different things by the same words, and that mismatch costs time. This dictionary flips the problem: instead of more abstraction, it translates the everyday terms of LLM-driven development into operational, decision-ready language you can act on during implementation and debugging.
What Sets It Apart
- Focus on actionable definitions: each entry explains not just what a term means but why it matters for engineering decisions (so what: you can choose a mitigation like compaction or a tooling change with confidence).
- Practical, example-led guidance: entries include typical failure modes and concrete fixes (so what: reduces trial-and-error when an agent drifts or hallucinate).
- Designed for harnesses and teams, not academia: covers system prompts, tools, sessions, and deployment trade-offs (so what: helps engineers diagnose failures that look like “model problems” but are harness or AX issues).
- Lightweight, browseable format with pointers to further resources (so what: quick to consult during a debugging session or design review).
Who It's For and Trade-offs
Great fit if you build or operate LLM-based developer tooling, agent harnesses, or integrate models into CI/CD: it shortens the learning curve for engineers who must reason about tokens, context budgets, and toolchains. Look elsewhere if you need low-level research mathematics or training recipes—the dictionary intentionally omits deep theoretical derivations in favor of engineering heuristics and operational patterns.
