Design AI assistants people can trust.
A free, hands-on course in conversation design and LLM engineering: how to choose a model, ground it in your data, give it tools, orchestrate it, test it and keep it safe. Rebuilt for how assistants are actually made in 2026.
- 24modules
- 10case studies
- 5 h 40 minof study time
- 100+languages (machine translation)
- Freeno sign-up
Pick your path
Every module stands on its own, but you probably don’t need all 24 first. Choose a role for a suggested route.
Conversation designer
You shape what the assistant says, when it asks, how it recovers and how it sounds. Focus on principles, prompting, fallbacks, specs, voice and journeys; skim the code.
About 3 h 2 min of study time, plus the knowledge checks.
- 1Foundations of conversational AI10 min
- 2From intents to LLMs14 min
- 3Conversation design for LLM assistants11 min
- 5Prompting across different models17 min
- 11Fallbacks and graceful degradation13 min
- 13Inbound and outbound assistants: campaigns and messaging15 min
- 15Writing test case specs for AI15 min
- 18Safety, security and ethics20 min
- 19Voice assistants12 min
- 20Voice localization: STT and TTS across languages14 min
- 21Multimodal journeys13 min
- 22Capstone project and next steps12 min
- 23Context engineering and memory16 min
Developer
You build the system. Focus on models, prompting, retrieval and ingestion, tools, orchestration, guardrails, fallbacks and evaluation, with working code.
About 3 h 22 min of study time, plus the knowledge checks.
- 2From intents to LLMs14 min
- 4The LLM landscape: choosing a model14 min
- 5Prompting across different models17 min
- 6Retrieval-augmented generation (RAG)12 min
- 7Preparing documents for ingestion18 min
- 8Tools, function calling and MCP16 min
- 9Orchestration: state machines, agents and multi-agent systems14 min
- 10Guardrails13 min
- 11Fallbacks and graceful degradation13 min
- 12Platforms and hybrid architectures9 min
- 14Evaluation and observability12 min
- 18Safety, security and ethics20 min
- 23Context engineering and memory16 min
- 24Evaluating agents14 min
QA and tester
You decide whether it is safe and good enough to ship. Focus on guardrails, fallbacks, evaluation, specs, the QA playbook, locale testing and release gates.
About 2 h 53 min of study time, plus the knowledge checks.
- 1Foundations of conversational AI10 min
- 2From intents to LLMs14 min
- 10Guardrails13 min
- 11Fallbacks and graceful degradation13 min
- 14Evaluation and observability12 min
- 15Writing test case specs for AI15 min
- 16A QA playbook for testing AI17 min
- 17Data, analytics and BI for assistants19 min
- 18Safety, security and ethics20 min
- 20Voice localization: STT and TTS across languages14 min
- 22Capstone project and next steps12 min
- 24Evaluating agents14 min
Data and BI analyst
You turn conversations into insight. Focus on the event schema, metrics, dashboards, experiments, campaigns, privacy and retrieval metrics.
About 2 h 26 min of study time, plus the knowledge checks.
- 1Foundations of conversational AI10 min
- 4The LLM landscape: choosing a model14 min
- 6Retrieval-augmented generation (RAG)12 min
- 7Preparing documents for ingestion18 min
- 13Inbound and outbound assistants: campaigns and messaging15 min
- 14Evaluation and observability12 min
- 17Data, analytics and BI for assistants19 min
- 18Safety, security and ethics20 min
- 22Capstone project and next steps12 min
- 24Evaluating agents14 min
Content and knowledge owner
You own the documents the assistant answers from. Focus on retrieval, document design and ingestion, fallbacks and the metrics that reveal content gaps.
About 1 h 43 min of study time, plus the knowledge checks.
Product or CX leader
You decide what to build and how to measure it. Focus on the landscape, architecture, guardrails, channels, campaigns, metrics and real incidents.
About 2 h 48 min of study time, plus the knowledge checks.
- 1Foundations of conversational AI10 min
- 4The LLM landscape: choosing a model14 min
- 9Orchestration: state machines, agents and multi-agent systems14 min
- 10Guardrails13 min
- 11Fallbacks and graceful degradation13 min
- 12Platforms and hybrid architectures9 min
- 13Inbound and outbound assistants: campaigns and messaging15 min
- 17Data, analytics and BI for assistants19 min
- 18Safety, security and ethics20 min
- 21Multimodal journeys13 min
- 22Capstone project and next steps12 min
- 23Context engineering and memory16 min
The curriculum
From first principles to production. Each module has worked examples, a knowledge check and sources you can verify.
- 01
Foundations of conversational AI
What an AI assistant actually is, what makes conversation hard to design, and how to tell whether one is working.
- 02
From intents to LLMs
How machines understand language, from intents and slots to embeddings and large language models, and when each approach still wins.
- 03
Conversation design for LLM assistants
Personas, system prompts as design documents, turn design, repair strategies and handoff: the craft that makes an assistant feel considered.
- 04
The LLM landscape: choosing a model
How to choose among proprietary and open models: criteria that matter for assistants, the tiers and routing strategies, and a dated snapshot of the families you will meet.
- 05
Prompting across different models
Why a prompt that works on one model fails on another, what actually differs, and how to keep prompts portable, versioned and tested.
- 06
Retrieval-augmented generation (RAG)
Ground your assistant in your own content: preparing knowledge, chunking, hybrid search, reranking, citations, evaluation and the failure modes to expect.
- 07
Preparing documents for ingestion
How to write, structure and manage source documents so retrieval works: single-document design, multi-document corpora, metadata, versions, format pitfalls and an ingestion pipeline with tests.
- 08
Tools, function calling and MCP
Let an assistant act: how tool calling works, how to design tools models can use safely, and what the July 2026 stateless MCP specification changes.
- 09
Orchestration: state machines, agents and multi-agent systems
Choose the right architecture: deterministic flows, a single tool-using agent or several cooperating agents, and how to combine them in a hybrid design.
- 10
Guardrails
Layered controls on inputs, retrieval, prompts, tools and outputs: what to enforce in code, what to ask of the model, how to tune false positives and how to test them.
- 11
Fallbacks and graceful degradation
Design what happens when understanding, retrieval, tools, models or channels fail: a fallback ladder, resilient calls, honest wording, handoff when no one is available and tests that break things on purpose.
- 12
Platforms and hybrid architectures
Cloud conversational services, open-source frameworks, visual builders and agent runtimes: how to choose, how to combine them, and how to migrate a legacy intent bot.
- 13
Inbound and outbound assistants: campaigns and messaging
Designing assistants that start conversations as well as answer them: consent and compliance, campaign and mailing design, two-way replies, voice outreach, architecture and the metrics that keep you honest.
- 14
Evaluation and observability
Build test sets, automated checks and LLM judges, then trace, monitor and gate every change so assistants improve instead of drifting.
- 15
Writing test case specs for AI
How to specify tests for assistants: fields, oracles, multi-turn scenarios, handling nondeterminism, coverage and review, with reusable templates.
- 16
A QA playbook for testing AI
How testers approach assistants: a test pyramid, exploratory charters, adversarial testing, flakiness, defect taxonomy, bug reports and release gates.
- 17
Data, analytics and BI for assistants
What to log, how to model it, which metrics to trust, how to build dashboards and experiments, and how to protect privacy while you learn from conversations.
- 18
Safety, security and ethics
The risks that matter for assistants: prompt injection, excessive agency, data exposure, honesty and accountability, mapped to the OWASP Top 10 for LLM applications.
- 19
Voice assistants
Design for the ear: voice architectures, latency budgets, barge-in, confirmation, repair and writing that sounds right when spoken.
- 20
Voice localization: STT and TTS across languages
Adapting speech recognition and synthesis to languages, dialects and domains: locales, vocabulary, normalization, SSML and lexicons, multilingual content, code-switching and testing per locale.
- 21
Multimodal journeys
Designing tasks that move across voice, chat, screens, images and documents: choosing the right modality, carrying context and identity, handling images and files, and testing cross-channel flows.
- 22
Capstone project and next steps
Apply everything: design, build, test and review a small assistant, with a brief, deliverables, a rubric and a plan for staying current.
- 23
Context engineering and memory
Prompt engineering writes the instructions. Context engineering decides everything the model sees at each step. Learn what to put in, what to keep out, and how to design memory that helps without leaking or misleading.
- 24
Evaluating agents
An agent is judged by what it did, not only what it said. Learn to grade outcomes, trajectories, tool use and whole conversations, calibrate model judges, use simulated users and gate releases.
Learn from real incidents
10 public cases, each traced to design decisions and what to do differently. All facts link to sources.
- 2024
When the chatbot's answer became the airline's problem
A tribunal held Air Canada responsible for wrong bereavement-fare advice from its own chatbot.
- 2024
The delivery chatbot that criticized its own company
After a system update, DPD's chat assistant swore at a customer and called DPD the worst delivery firm. The company switched off the AI part the same day.
- 2024
The city chatbot that told businesses to break the law
New York City's MyCity chatbot gave businesses incorrect and illegal guidance. The city added disclaimers, then shut it down in 2026.
- 2025
The assistant that handled two-thirds of chats, and the lesson about quality
Klarna's AI assistant took on a huge share of customer chats. Then the CEO said cost had been weighted too heavily, and the company began hiring humans again.
- 2025
The support bot that invented a policy to explain a bug
Cursor's AI support agent told a user that one device per subscription was a security policy. No such policy existed, and customers cancelled.
- 2024
The voice-ordering system that was switched off after viral mistakes
McDonald's ended its automated drive-thru ordering test with IBM after videos of order errors spread, and reports pointed to trouble with accents and dialects.
- 2025
The coding agent that deleted a production database during a code freeze
During a public experiment, an AI coding agent ran destructive commands against a live database despite a freeze, and then produced misleading output.
- 2024
When regulators said AI voices on the phone count as artificial voices
In February 2024 the US FCC ruled that AI-generated voices are artificial under the TCPA, bringing consent rules to AI outbound calls.
- 2025
One tool, three legs: how an issue tracker became a data leak
A single MCP tool could read public issues anyone can file, reach private repositories and open pull requests. Together, that is a path for stealing data.
- 2026
A clinic's chatbot invented medical titles, and a court said they were the clinic's own words
A website chatbot told visitors that two doctors held specialist titles, two of which do not exist. A German appeals court held the company responsible for the answer.
One-page cheat sheets
19 printable references, also available as PDF: prompts, RAG, tools and MCP, guardrails, test specs, metrics, voice, campaigns and more.
Quick guides
Practical, dated answers to the questions people ask most.
Chatbot, assistant, agent or copilot? A plain-English guide
The four words are used loosely. Here is what each one means in practice, how much it is allowed to do on its own, and how to choose.
GuideAI disclosure and chatbot regulation: a tracker for designers
What the EU AI Act now requires of chatbots and voice assistants, what was delayed and what was not, and a design checklist. Dated and sourced. Not legal advice.
GuideFrequently asked questions
48 short answers on design, agents, evaluation, safety, rules, voice and careers.
What changed
Every release is logged here, newest first. Earlier releases stay below, unchanged. See the full log.
November 3, 2026Release 2026.11.03dFull content check: corrections8 changes
Every page added or changed since the demand research was read against its sources. No factual errors were found. The corrections below are wording and consistency.
Updated (6)
- UpdatedModule 23: memory policy and code comment
The memory policy table now says that its consent and deletion rows become test cases, a note explains that memory about a person is personal data under GDPR and LGPD, and the comment in the code example now matches what the code does.
- UpdatedModule 24: what to grade
The section title said three things but its table lists five. It now reads "What to grade, and at what level". The wording on simulated users is softened to "often" and "tend to", because that behavior varies.
- UpdatedGuide: Messaging and WhatsApp, wording
The caution about the Brazil status is now neutral text instead of first person, and the message-template rule states the usual 24-hour window.
- UpdatedGuia (português): WhatsApp, redação
A mesma cautela sobre o Brasil agora está em texto neutro, e a sugestão de módulos virou links ("Comece pelos módulos 3, 19 e 18").
- UpdatedFAQ: simulated users
The answer now says simulated users tend to be more cooperative than real ones.
- UpdatedGlossary: simulated user
Now says "often kinder than real users".
Notes (2)
- NoteWhat was checked
The two new modules and their quizzes, the 48 FAQ answers, the regulation tracker, the messaging and Portuguese guides, the explainer, career and human-handoff guides, five cheat sheets, both new case studies, the new sections in modules 8, 17, 18 and 19, and the 25 new glossary terms. The note that MCP became stateless (specification of 28 July 2026) was confirmed against the MCP project's own release post, and the GitHub case study against Simon Willison's original article.
- NoteWhat was not re-read
Modules 1 to 22 as they stood before the demand research, and the eight earlier case studies, were not read again in this pass.
November 3, 2026Release 2026.11.03cLiability, human handoff and a Portuguese starting point6 changes
The last items of the demand research: what courts and lawmakers now say about who owns a chatbot's words, a plain list of what should stay with a person, and a Portuguese page that maps the course.
New (3)
- NewCase study: A clinic's chatbot invented medical titles
A German appeals court (Higher Regional Court Hamm, 12 May 2026) held a company responsible for false answers its website chatbot gave about its doctors' titles. What the court said, what it did not, and how to design so a bot cannot make claims about your organization. Not final.
- NewGuide: What should stay with a person
A list of ten situations an AI agent should not handle alone, what the assistant does instead, four questions to ask before automating a task, how to hand off well and what to test.
- NewGuia (português): Design de IA conversacional, por onde começar
A map of the course in Portuguese, answers to three common questions (chatbot, assistente ou agente; aviso de IA; quem responde por um erro) and a pointer to the WhatsApp guide.
Updated (2)
- UpdatedModule 18: liability watch
Adds a dated box with the Hamm ruling and California AB 316, which stops a defendant from arguing that the AI acted autonomously, and a pointer to the guide on what stays with a person.
- UpdatedFAQ: two answers updated, two new (48 in total)
The answers on which calls an agent should never handle alone and on liability now point to the new guide and case study. New questions: whether an "AI can make mistakes" disclaimer protects a company, and what California AB 316 changes.
Notes (1)
- NoteBoth legal items were checked against primary sources
The Hamm ruling against the court's own press release and the published full text; AB 316 against the enacted bill (Chapter 672, Statutes of 2025). Both pages say these are summaries, not legal advice, and that the Hamm ruling is not final.
November 3, 2026Release 2026.11.03bFollow-up: fact check3 changes
A second pass on the same day: the new pages were spot-checked against the research sources.
Notes (3)
- NoteFact check: regulation tracker
The Article 50 dates, the Digital Omnibus postponement and the 2 December 2026 marking grace were re-checked against several law-firm summaries and found consistent. The California and FCC references rest on earlier, widely documented rulings and were not re-researched. No change to the text was needed.
- NoteFact check: messaging guide
The guide already says that reports disagree on the Brazil status and that it was not checked against Meta's own terms. That caution was kept.
- NoteWhat was not re-checked
Pages not named in this release were not independently re-verified in this pass, and no vendor statistics were added.
November 3, 2026Release 2026.11.03Demand-driven update: rules, agent security, evaluation, memory and messaging27 changes
A research pass on what people search for around conversational design, AI bots and agentic support found gaps in regulation, agent evaluation, context and memory, agent security, support metrics and messaging channels. This release fills them and adds guides for the most common questions.
New (15)
- NewModule 23: Context engineering and memory
What goes into a model's context, how long conversations fail, and how to design memory for support (what to remember, what never to, consent, forgetting, memory poisoning).
- NewModule 24: Evaluating agents
Trajectory, tool-use and state checks, multi-turn quality, calibrating LLM judges, simulated users, a 50-example starter set and evals in CI.
- NewGuide: Chatbot, assistant, agent or copilot?
A plain-English comparison and an autonomy ladder, for the most common beginner question.
- NewGuide: AI disclosure and chatbot regulation tracker
EU AI Act Article 50 (applicable since 2 August 2026), what the Digital Omnibus delayed and did not, and a design checklist. Dated, with sources, not legal advice.
- NewGuide: Frequently asked questions (46 answers)
Short answers grouped by topic, each linking to the page with the detail.
- NewGuide: Becoming a conversation designer
What the role asks for in 2026 job ads, how the modules map to the skills, and how to build a portfolio from the capstone.
- NewGuide: Messaging and WhatsApp for support
WhatsApp's 2026 policy on general-purpose bots, voice notes, consent, handoff continuity and what to verify.
- NewGuia (português): Agentes de IA no WhatsApp
Versão curta do guia de mensageria, escrita em português para o público brasileiro.
- NewCase study: One tool, three legs
How a single MCP tool combined private data, untrusted content and an outbound channel, and what it teaches about support agents.
- NewCheat sheet: Lethal trifecta test (10 minutes)
Three questions to ask about every agent, and how to break one leg.
- NewCheat sheet: Agent evaluation checklist
Datasets, graders, trajectory and state checks, and release gates.
- NewCheat sheet: Resolution, deflection, ROI and pricing
Definitions that cannot be gamed, a cost-per-resolution formula and red flags in vendor benchmarks.
- NewCheat sheet: Agent protocol map
MCP, A2A and the commerce protocols as layers, with a note on what is still changing.
- NewCheat sheet: Voice go-live checklist
Latency, barge-in, entity accuracy, consent, escalation and cost checks before a voice agent takes real calls.
- NewGlossary: 25 new terms (70 in total)
Context engineering, trajectory, lethal trifecta, tool poisoning, deflection rate, resolution rate, memory types, end-of-turn detection and more.
Updated (10)
- UpdatedModule 18: lethal trifecta, indirect injection, tool poisoning, policy promises
New sections on the three-capability test, attacks that arrive through tickets and forms, tool-description poisoning, and why a bot must not invent commitments. Adds one quiz question and three sources.
- UpdatedModule 8: protocol map and tool poisoning
New section placing MCP, A2A and the commerce protocols in layers, plus a tool-poisoning note in the security section. Adds one quiz question.
- UpdatedModule 17: resolution, deflection, cost per resolution, ROI, pricing
New sections on metrics that cannot be gamed and on outcome-based versus per-conversation pricing. Adds one quiz question.
- UpdatedModule 19: entity accuracy and unit economics
New sections on getting names, numbers and codes right and on what a minute of voice really costs. Adds one quiz question.
- UpdatedModule 1: link to the chatbot, assistant, agent and copilot guide
One pointer added.
- UpdatedModule 9: pointer to the context engineering module
One pointer added after the memory and state section.
- UpdatedModule 14: pointer to the agent evaluation module
One pointer added.
- UpdatedModule 13: pointer to the messaging guide
One pointer added.
- UpdatedModule 22: next steps now include modules 23 and 24 and the guides
The capstone's closing section lists where to go next.
- UpdatedCheat sheet: pre-launch safety checklist now covers launch and incident readiness
Added a kill switch, rollback, human override, incident log and an owner for every policy.
Notes (2)
- NoteFacts that move fast carry a "last verified" date
The regulation tracker, the messaging guide and the protocol sections say when they were checked and against what.
- NoteWhat was left out
Vendor statistics (resolution rates, ROI, salaries) and one list of incidents that could not be confirmed elsewhere were not used. Rulings and policies that rest on a single secondary source were kept out or marked to verify.
October 27, 2026Release 2026.10.27Rebuild since the first edition5 changes
The field moved fast. This rebuild reflects it, and each fast-moving page is dated.
Frequently asked questions
Is the course free, and do I need to sign up?
Yes, it is free and open to everyone. There is no account. Your progress is saved in your own browser on your device.
Who is this course for?
Conversation designers, developers, QA engineers and testers, data and BI analysts, and product or customer-experience leaders. Each role has a suggested path through the modules.
Do I need to know how to code?
No. Designers, QA, data and product readers can follow most of it without coding. Developers get working Python examples, including a tool-calling loop, an MCP server and a test runner.
How long does the course take?
The estimated study time is shown on every module and on the curriculum page. Plan for the reading plus the knowledge checks and the capstone exercises.
Which AI models does it cover?
It is model-agnostic. Module 4 gives you criteria and a dated snapshot of the main proprietary and open-weight families, and module 5 shows how prompts differ across models. Specific versions change often, so the course links to official sources.
Does it cover RAG, MCP and agents?
Yes. It covers retrieval-augmented generation, tool use and the Model Context Protocol (including the 2026-07-28 revision), and how to choose between state machines, single agents and multi-agent systems.
How does it teach testing and analytics?
There are dedicated modules on evaluation, writing test case specs, a QA playbook for AI, and data and BI for assistants, with templates, SQL examples and release-gate checklists.
Is it available in my language?
The course is written in English. You can read it in over 100 languages with the built-in machine-translation menu in the header, and code stays untouched. Machine translation can be imperfect, so use the original English for exact wording.
Are there cheat sheets?
Yes. There are one-page cheat sheets for prompts, RAG and ingestion, tools and MCP, guardrails and fallbacks, test case specs, QA, metrics, voice, campaigns and safety. View them online, print them or download the PDF.
How do I track my progress?
Answer the knowledge check at the end of each module. The progress widget saves what you have read and your quiz results, shows your next step, and lets you export your progress as a file.
How current is the content?
Fast-moving pages carry a review date, and facts link to official or primary sources. The course was reviewed on October 27, 2026.
Nelson Abacherli Kawakami
I have spent more than two decades in design and digital innovation, and almost a decade of that specializing in conversational design: chatbots, IVRs and AI systems. I have also led creative and technical teams for 13 years. This course comes from that mix of design thinking and hands-on engineering.