Browse tutorials
Tutorials
Shipping an AI System: Architecture, Observability, Cost, and Reliability
Package an evaluated AI workflow behind a local API boundary with structured logs, latency accounting, fallbacks, privacy controls, and rollback thinking.
Securing AI Applications: Prompt Injection, Data Leakage, and Tool Safety
Build a harmless local prompt-injection test lab and map trust boundaries, least privilege, output validation, sandboxing, approvals, and incident response.
Evaluating Generative AI: From Vague Impressions to Reproducible Tests
Create a small golden dataset and evaluation harness for groundedness, citations, retrieval, task success, regression, and uncertainty.
Agent Memory: What AI Systems Should Remember—and What They Should Forget
Implement short-term state, episodic records, and retrieval-based memory while handling retention, stale facts, conflicts, privacy, and deletion.
AI Agents: Models, Tools, State, and the Execution Loop
Build a bounded tool-using agent with explicit state, permissions, retries, and stopping rules, then place it inside an operational investigation workflow.
Beyond Basic RAG: Hybrid Search, Reranking, and Better Retrieval
Compare lexical, dense, hybrid, and reranked retrieval on a small labelled set, then decide when the extra complexity is justified by measurable search failures.
Retrieval-Augmented Generation: Giving Language Models Verifiable Context
Build a small citation-aware retriever, inspect ranked passages before generation, and learn why chunking, permissions, freshness, and evaluation matter more than adding context blindly.
Tool-Using AI: From Free-Form Text to Reliable Actions
Build a provider-agnostic tool dispatcher that validates arguments, enforces an allow-list, executes deterministic functions, and handles retries without treating generated text as permission.
Context Engineering: Designing the Information an AI System Actually Sees
Treat instructions, examples, retrieved evidence, tools, constraints, and output schemas as one context system, then test a local structured-extraction pipeline under a fixed budget.
Fine-Tuning Language Models: Full Training, LoRA, and the Practical Middle Ground
Understand what full fine-tuning changes, how adapters and LoRA reduce trainable parameters, and how to choose adaptation only after a baseline and held-out evaluation.