Back to index
Other11 short video lessons, a few hours; one-hour 28-question certification exam·Course free; Developer Edition licence free up to 1,000 conversations a month; production licence priced by sales

Conversational AI with CALM and the Rasa Pro Developer Certification

3.9

The clearest free explanation I know of how to put guardrails around an LLM-driven assistant using explicit flows. The catch is that the open source framework most people remember is in maintenance mode and what you are really learning is Rasa Pro.

What We Liked

  • CALM is a genuinely useful idea: the LLM interprets the user, flows decide what the business does
  • Course is free and short, with video for every lesson
  • Covers the unglamorous parts: repairing conversations that go off the happy path, custom actions, running in production
  • Free Developer Edition licence lets you build real assistants while you learn
  • Industry starter packs for banking, telecom and insurance show realistic flows

What Could Be Better

  • Rasa Open Source is in maintenance mode, so the skills tie you to a commercial product
  • Production licences are not priced publicly and third-party estimates start in the tens of thousands a year
  • Exam is 28 multiple-choice questions, which proves recall more than build ability
  • A large archive of pre-CALM courses sits alongside the new material and is easy to wander into by mistake

Detailed review

Rasa used to be the default answer to the question of how to build a chatbot without handing your data to a big cloud vendor. That has changed. Rasa Open Source is now in maintenance mode and active development happens in Rasa Pro, which is commercial. There is a free Developer Edition licence, capped at 1,000 conversations a month (100 if the users are your own employees), and beyond that you talk to sales. The course that matters now is Conversational AI with CALM on the Rasa Learning Center.

It runs through 11 short video lessons: what CALM is, how an assistant is structured, setting up your environment, writing flows for business logic, dialogue understanding, custom actions, handling deviations from the happy path, enterprise search, and running in production. The idea behind CALM is the part I would recommend to anyone. You let the language model do what it is good at, working out what the user means, and you keep the actual decisions in explicit flows that a developer can read and test. That is the right instinct for a bank or an insurer, where an assistant inventing a refund policy is a real liability. The lesson on repairing conversations, when a user changes their mind or goes off on a tangent, is the most practical material I have seen on the topic from any vendor. The certification exam is short: 28 multiple-choice questions, one hour, and you need more than 85% to pass.

You get a downloadable certificate straight away. I could not find a clear exam fee on the Learning Center, and it is not the kind of test that proves you can build something. Rasa's own advice is that the best preparation is building an assistant, and I agree. One annoyance: the site still hosts a big archive of older courses on intents, stories and the classic NLU pipeline. They are clearly labelled as archive, but search engines send people to them, and they teach an approach Rasa itself has moved away from. Search interest is small, around 30 searches a month in the US for Rasa academy, though it jumped recently. My take: take the CALM course for the design pattern, which applies equally to Google's flows, Voiceflow or a homegrown agent.

Treat the certificate as a nice extra if you are already on Rasa Pro, not a reason to adopt it.

[ final ]

The verdict.

Worth a few hours for anyone building customer-facing assistants in a regulated business, even if you never use Rasa, because the design pattern transfers. Only chase the certificate if your employer already runs Rasa Pro.