Short, focused courses on retrieval, reasoning, and production AI — built for engineers who want to go beyond the tutorials.
Query transformation, sparse and dense retrieval, reranking, fusion, chunking, and production patterns — the full retrieval stack.
System topology, Pub/Sub, Cloud Functions, and API Gateway, plus foundation model access, agent orchestration with Agent Builder and ADK, RAG infrastructure, training and serving, and evaluation and safety on Vertex AI.
Evals, tracing, regression testing, and monitoring LLM systems in production — the discipline that separates demos from products.
Threat modeling, input/output guardrails, training-time safety, red teaming, safety evaluation, production architecture, agent isolation, and governance — defense-in-depth for LLM systems.
Supply-chain compromise, MCP's rug-pull risk, the lethal trifecta and Rule of Two, production gateways with audit trails and SSO, daily mitigation habits, and detection and incident response — securing the agent stack that model-safety guardrails were never built to reach.
Dialog structure, error recovery, agency calibration, memory architecture, and measurement, plus proactive agents, trigger design, ambient interaction, and trust calibration — the interaction-layer patterns that determine whether users trust and return to an AI-powered product.
Token-cost governance, reasoning-model tradeoffs, agent and MCP security beyond guardrails, and hardening MCP in production with gateways, audit trails, and SSO — the operational edge of running AI systems in 2026.
Multi-agent review panels, checkpoint-based epic planning, mutation and property-based testing, and the judgment calls behind a production-grade Claude Code plugin — how to build engineering workflows that hold up under real work.
From bag-of-words and static embeddings through attention, transformer architectures, tokenization, fine-tuning, production NLP tasks, evaluation, and efficient inference — the architecture choices behind modern language systems.
Working vs. long-term memory, extraction-based vs. temporal-graph architectures, the token economics of remembering, and the write-consolidate-forget lifecycle — how agents hold onto what they learn about you across sessions.
LoRA and QLoRA adapters, the wider PEFT method landscape, DPO and preference alignment beyond supervised fine-tuning, evaluation, and production deployment pitfalls — when fine-tuning earns its cost and how to do it efficiently.
Quantization fundamentals, GGUF and local serving formats, the small-model landscape, distillation and specialization, and production economics — running capable models on constrained hardware without a frontier-scale bill.
Article 50 disclosure and deepfake-labeling obligations, GPAI provider requirements, and practical compliance steps for engineering teams — what actually changed, what didn't, and what's still coming.
Cycles and conditional branching, persistence and checkpointing, human-in-the-loop interrupts, and multi-agent supervisor/swarm patterns — when a graph-based orchestration layer earns its keep over a simpler chain.
Delta Lake, MLflow experiment tracking, Unity Catalog governance, distributed compute with Spark, and model serving — the lakehouse platform where AI engineering meets data infrastructure.
The four core RAG metrics, synthetic test-set generation, and turning evaluation scores into concrete fixes and CI gates — evaluation built specifically for retrieval-augmented systems.
CoT, self-consistency, tree of thoughts, and structured reasoning — when and how to use each to improve LLM output quality.
Tool use, planning, memory, multi-agent coordination, and evaluation — everything you need to ship agents that actually work.