I finished AI Patterns. I can call models, parse JSON, hold conversations, batch and retry. What's left?
The patterns that turn a demo into something you'd ship. Embeddings — turning text into vectors so you can compare meaning, not strings. RAG — grounding answers in your own data instead of model knowledge. Eval suites at scale — knowing when a prompt change improved or regressed. Cost & latency — model routing, caching, parallel calls. Guardrails — PII redaction, rate-limit handling, fallback chains. Human-in-the-loop — gating risky actions on approval. Recovery — multi-step agents that fall back when a tool breaks.
That sounds like ten different things.
It is. Each is one lesson, one concept, one minimal demonstration. The final week ties them together: a small RAG-with-eval-and-guardrails script on a tiny generic corpus. The skill is composition — these primitives plug into anything you already know how to build.
And week 4?
Production observability — prompt versioning, A/B prompt tests, cost dashboards, guardrail composition, agent traces, a final synthesis. The v1 north star is here: by the end, you can compose Python + Automation + AI primitives onto any task.
AI Mastery is the capstone of the v1 arc.
Week 1 — Embeddings and RAG. Cosine similarity, embedding API shape, chunking, storing embeddings, semantic search, the full RAG pipeline.
Week 2 — Quality and optimization. Citations, RAG failure modes, eval suites at scale, model routing, caching, parallel calls.
Week 3 — Production guardrails. PII detection, rate-limit-aware retry, fallback chains, human-in-the-loop approval (Gmail-self-send), multi-step agents with recovery, full RAG-agent synthesis.
Week 4 — Observability and shipping. Prompt versioning, A/B prompt testing, cost dashboards (write usage to a Sheet), guardrail composition, agent observability, final integration synthesis.
Write production-shaped scripts that ground answers in your data, catch regressions before shipping, optimize cost and latency, redact PII before sending, fall back gracefully when a tool fails, and require human approval before risky actions. The full v1 kit — Python primitives + automation tools + LLMs — composed onto any task.
AI Patterns (or equivalent fluency with Agent(model), output_type, tool calling, multi-turn). Python Patterns (file I/O, JSON persist, helpers, decorators). Automation Foundations (Composio helpers — needed for HITL and the synthesis). Max-tier subscription.
Rate each statement on the 1-5 scale. The same six prompts return on day 30.
I finished AI Patterns. I can call models, parse JSON, hold conversations, batch and retry. What's left?
The patterns that turn a demo into something you'd ship. Embeddings — turning text into vectors so you can compare meaning, not strings. RAG — grounding answers in your own data instead of model knowledge. Eval suites at scale — knowing when a prompt change improved or regressed. Cost & latency — model routing, caching, parallel calls. Guardrails — PII redaction, rate-limit handling, fallback chains. Human-in-the-loop — gating risky actions on approval. Recovery — multi-step agents that fall back when a tool breaks.
That sounds like ten different things.
It is. Each is one lesson, one concept, one minimal demonstration. The final week ties them together: a small RAG-with-eval-and-guardrails script on a tiny generic corpus. The skill is composition — these primitives plug into anything you already know how to build.
And week 4?
Production observability — prompt versioning, A/B prompt tests, cost dashboards, guardrail composition, agent traces, a final synthesis. The v1 north star is here: by the end, you can compose Python + Automation + AI primitives onto any task.
AI Mastery is the capstone of the v1 arc.
Week 1 — Embeddings and RAG. Cosine similarity, embedding API shape, chunking, storing embeddings, semantic search, the full RAG pipeline.
Week 2 — Quality and optimization. Citations, RAG failure modes, eval suites at scale, model routing, caching, parallel calls.
Week 3 — Production guardrails. PII detection, rate-limit-aware retry, fallback chains, human-in-the-loop approval (Gmail-self-send), multi-step agents with recovery, full RAG-agent synthesis.
Week 4 — Observability and shipping. Prompt versioning, A/B prompt testing, cost dashboards (write usage to a Sheet), guardrail composition, agent observability, final integration synthesis.
Write production-shaped scripts that ground answers in your data, catch regressions before shipping, optimize cost and latency, redact PII before sending, fall back gracefully when a tool fails, and require human approval before risky actions. The full v1 kit — Python primitives + automation tools + LLMs — composed onto any task.
AI Patterns (or equivalent fluency with Agent(model), output_type, tool calling, multi-turn). Python Patterns (file I/O, JSON persist, helpers, decorators). Automation Foundations (Composio helpers — needed for HITL and the synthesis). Max-tier subscription.
Rate each statement on the 1-5 scale. The same six prompts return on day 30.
Create a free account to get started. Paid plans unlock all tracks.