hub.gazar.dev · course

Production-Ready Systems with LLMs and Agents

Build LLM and agent systems that survive real traffic, cost, and failure. The part the framework tutorials skip: the architecture decisions that make LLM and agent systems hold in production. Twelve 90-minute live sessions and five production-grade artifacts, over six weeks, and every week ships with a slide deck, a cheat sheet, and a hands-on coding lab. maven.com/gazar/production-ready-systems-with-llms-and-agents · Companion code (TypeScript + OpenAI): github.com/ehsangazar/maven-llms-and-agents-6-weeks

Every lab ships as runnable code

TypeScript and OpenAI, behind one provider seam you can swap. Clone the companion repo, or jump to a lab's code under its week below.

github.com/ehsangazar/maven-llms-and-agents-6-weeks ↗

A field guide to the tools

Each week below ends with the common tools for that stage of the stack. Every tool has a decision-first page: what it is, the one job it does, when to reach for it, when to skip it, and where it fits your labs. The tools are a means; the architecture is the lesson.

Pre-course · set the target

Before Jul 13

Week 1 · Foundations: workflows, agents & the code/model boundary

Jul 13 – 19

Week 2 · Context engineering & retrieval

Jul 20 – 26

Field guide · tools for this week

Week 3 · Cost, latency & reliability

Jul 27 – Aug 2

Field guide · tools for this week

Week 4 · Agent architecture & security

Aug 3 – 9

38 slides

S7 · Agent patterns and their failure modes

The loop and its four stops, tools as typed contracts, MCP, three ways to plan, the compounding maths that sets your step cap, durable runs that resume instead of restarting, and single versus multi-agent. Every pattern mapped to how it breaks, with runnable TypeScript for each. Tue Aug 4.

/llms-and-agents/s07-agent-architecture
34 slides

S8 · Securing agents + choose your architecture

Why prompt injection cannot be prompted away, the three shapes of the attack, the lethal trifecta, and zero-click exfiltration. Then five layers you can run: the trust boundary, scoped capabilities and taint, approval gates, egress allow-lists, supply-chain pinning, and a red-team corpus. Workshop: pick your architecture and threat-model it. Thu Aug 6.

/llms-and-agents/s08-securing-agents
artifact Project 3 · due Sun Aug 9

P3 · Agent Architecture Decision + Threat Model

The pattern you chose, why, and the threats it exposes with the mitigations you'll rely on.

/llms-and-agents/p3-agent-architecture-threat-model
cheat sheet Reference · cheat sheet

Agent Patterns & Prompt-Injection Cheat Sheet

Agent anatomy, tools as contracts, the ReAct loop, single vs multi-agent, prompt injection (OWASP LLM01), defense-in-depth, the trust boundary, and real cases.

/cheatsheets/agent-security
hands-on lab Lab 4 · build it

Build a Tool-Using Agent with Guardrails

A ReAct agent with least-privilege tools, a step cap, an approval gate before writes, and an injected document it must refuse. Starter code + a threat model.

/llms-and-agents/lab-guardrailed-agent

Week 5 · Evals & observability

Aug 10 – 16

27 slides

S9 · Trajectory evals, not just outputs

Why grading the final answer isn't enough for agents, told through the chatbot ruling that made an airline liable. Three blocks: what to assert about a trajectory (required, forbidden, budget), who grades the fuzzy half and how to score the judge against its own baseline, and when a result blocks a release. Offline sets, online sampling, and the loop where failed production traces become permanent cases. Tue Aug 11.

/llms-and-agents/s09-trajectory-evals
28 slides

S10 · Workshop: harness, tracing & runbook

The span recorder and the replay test written as code, the three dashboards and why a mean and an averaged p95 both lie, content-hashed prompt versions, and where a human stays. Finish the failure-mode runbook. Thu Aug 13.

/llms-and-agents/s10-harness-tracing
artifact Project 4 · due Sun Aug 16

P4 · Eval Harness Plan + Failure-Mode Runbook

How you'll catch regressions before users do, and the runbook completed for when something still gets through.

/llms-and-agents/p4-eval-harness-plan
cheat sheet Reference · cheat sheet

Evals & Observability Cheat Sheet

Output vs trajectory evals, the three graders, the four metrics, offline vs online, the flywheel, an eval-case and a trace in JSON, and the three dashboards.

/cheatsheets/evals-observability
hands-on lab Lab 5 · build it

Build an Eval Harness

A golden set, a runner, rule + LLM-judge graders, a trajectory check, and a pass-rate report that catches a regression you deliberately introduce. Starter code included.

/llms-and-agents/lab-eval-harness

Week 6 · Capstone: design, present, defend

Aug 17 – 23

Field guide · pull it together

Small-group intensive. Each Tuesday teaches the concept and its trade-offs; each Thursday is a workshop where you apply it to your own system and build that week's slice of a production-grade artifact. Taught by Ehsan Gazar, Principal Engineer, 16 years in production systems, 500+ mentees.