> cat ai_skills.txt
AI & LLM Engineering Skills
As LLMs become part of real products, they need the same rigor as any other backend — verified quality, safety and cost, not just a nice-looking demo. Here's what that looks like in practice.
Hallucination & Groundedness Testing
Building automated checks that verify an AI's answer is actually grounded in the facts it was given — not invented, no matter how confident it sounds.
✓ Proven in AI & LLM Testing FrameworkPrompt Injection & Guardrail Testing
Testing that a model safely refuses harmful or manipulative prompts — while making sure it doesn't over-refuse and block legitimate questions too.
✓ Proven in AI & LLM Testing FrameworkStructured Output Validation
Verifying AI-generated JSON actually parses and matches an exact expected schema — critical the moment an LLM's output feeds into other code.
✓ Proven in AI & LLM Testing FrameworkLatency & Token-Cost Budgeting
Asserting AI responses stay within acceptable speed and cost limits — treating an LLM call like any other API that has a performance budget.
✓ Proven in AI & LLM Testing FrameworkTest Automation for AI Systems
Building PyTest frameworks around a deterministic mock LLM client, so AI behavior can be tested reliably and cheaply in CI — no flaky outputs, no per-run API cost.
✓ Proven in AI & LLM Testing FrameworkPrompt Engineering & Test Data Design
Designing prompt sets as structured, reusable test data kept separate from test logic — covering both expected-good and deliberately adversarial model behavior.
✓ Proven in AI & LLM Testing FrameworkAutonomous Agent Skill Design
Writing structured operating guidelines that let an AI coding agent work autonomously within a defined domain — debugging, refactoring and validating its own changes safely, with clear scope and boundaries instead of a prompt for every step.
✓ Applied in real autonomous engineering workflowsSee it in code
These aren't abstract concepts — every skill above is backed by a real, working test suite with 12 passing tests and a CI pipeline that runs it on every push.