Kilevel Academy trains and examines the people who design, deploy and operate Kilevel in production. Each credential names exactly what its holder was assessed on.
Certified Kilevel experts, in your team or at your side, make every next deployment simpler and faster to scale.
A curriculum, the delivery method it teaches, and the people who are held to it.
The certifications themselves. Role based, assessed in a real tenant rather than on a question paper, and valid for a fixed period because the platform underneath them moves.
The implementation method every credential is assessed against: how a project is scoped, how a Flow is structured, what has to be tested, what has to be observable, and what a production handover contains.
Everyone who holds a credential, whether at Kilevel, at a customer or at a delivery partner, together with the instructors who train them. A public verification page stands behind every name.
It gives a procurement file a sentence it can use: the solution was implemented according to the Kilevel Deployment Standard by Kilevel-certified professionals. That is a different claim from “our supplier attended a training course”, and it is the claim the programme exists to make true.
Each is entered from the one below it, and each names what its holder was assessed on and how long that is good for.
For the people who run what is already live: operations teams, contact centre managers, customer administrators.
Access and resources: Users, roles, permissions and the shared resources a tenant reuses across projects.
Watching production: Monitoring runs and calls, reading an escalation, telling a configuration problem from a provider one.
Releases: Versioning, staged rollout and the rollback path, which is the part that gets skipped.
For the people who design and maintain conversations. The first credential most customer teams take, and the entry point to the professional track.
Structure over wording: State graphs, prompt modules, variables and the difference between a rule that holds and a rule the model can talk itself out of.
Tools and outputs: Tool calls, structured outputs, and what the conversation does while a call is still in flight.
The unhappy paths: Timeouts, missing data, interruptions, no slot available, a caller who wants a person. Assessed on these, not on the happy path.
For technically capable implementers working with Automations, APIs, data and the systems on the other side of them.
Integrations: HTTP and OAuth, asynchronous callbacks, RPA and the external systems an enterprise actually runs on.
Failure and retry: Idempotency, retry policy, duplicate execution and the race conditions that only show up under callback load.
Credentials and visibility: Secret handling, least privilege, and making a failure path observable instead of silent.
The main professional credential. Someone who can take a Kilevel solution from a first conversation to production on their own.
Discovery and architecture: Turning a business problem into a scope, a system design and an honest statement of what it will not do.
Evaluation: Building the test set before the rollout, and reading the result as a number rather than as an impression.
Production operations: Cutover, monitoring, the first bad week, and measuring the business outcome that justified the project.
An advanced credential with prerequisites and real delivery behind it. Assessed by a review board, not by a question paper.
Estate level design: Multi-system architecture, tenant isolation, integration strategy and where the boundaries of the platform sit.
Availability and capacity: Telephony and provider architecture, concurrency ceilings, failure planning and what each of them costs.
Defended, not submitted: The design is presented and questioned in person. An architecture you cannot defend out loud is not one you can hand over.
A designation, not an examination. It recognises people who have taken customer deployments into production and stayed accountable for them.
Certification first: An active AI Deployment Engineer credential is the floor, not the qualification.
Evidence from production: At least one supervised deployment, owned from discovery through to live operation, submitted as evidence and reviewed by an engineering panel.
Kept by delivering: The designation lapses without continued delivery. It describes what someone is doing, not what they once passed.
A permanent technical credential on a platform that ships every week would be worth nothing after the second year.
A delta module covering what changed, plus a mandatory module whenever a security or architecture change makes one necessary.
Every credential carries an ID and a public page: what was assessed, when it was issued, when it expires, and whether it is still current.
Kilevel AI generates exercises, simulates callers and marks competency gaps. The pass or fail on an advanced credential is made by a human examiner.
The programme is designed in alignment with the principles of ISO/IEC 17024 for bodies certifying persons: defined competencies, impartial assessment, records, recertification and human control over AI-supported marking. That is how it is built. It is not an accreditation, and we do not describe it as one.
Preparation and instruction are the cheap part. The credential is decided in a live tenant.
Self-paced preparation
Roughly eight to twelve hours before anyone travels: platform architecture, conversation and prompt engineering, tools and structured outputs, testing, failure handling, privacy and governance.
Two instructed days
Short conceptual sessions, guided builds, progressive exercises, group debugging of somebody else's work, enterprise case studies and design review. Attendance is not the credential.
A timed practical lab
An isolated tenant with a half-finished enterprise project in it, real requirements, existing components, several deliberately broken configurations and API mocks. Build it, test it, fix it, hand it over, and explain the design in writing.
The submitted Flow is run against cases the candidate never saw: an interruption during a tool call, an identification service that answers late, an API timeout, a caller switching language, a model reaching for something it does not know. It passes when it behaves safely across all of them.
Advanced candidates are handed an implementation that appears to work and is quietly wrong: a tool that executes twice, a retry with no idempotency, an exposed credential, a failure path nobody can see, an escalation that never fires. They have to find it and repair it.
Not whether the Flow runs. Task completion, resistance to invention, recovery behaviour, latency under load, security, maintainability and whether the thing it was built for actually happened. The same measures the Deployment Standard asks of a live project.
An easy credential is worse than none: it puts a Kilevel badge on work nobody checked. The examination is the same for a paying enterprise customer as for a Kilevel engineer, and a written explanation of the design is required so that nobody passes by trial and error alone.
The same credential means the same thing whether the holder is on your payroll or ours.
Internal: Your own engineers, certified and working inside your tenant on your terms. They keep the context that an external consultant rebuilds at the start of every engagement, and they are there on the Tuesday something breaks.
External: A certified engineer from Kilevel or from an authorised delivery partner, reaching into the same tenant under scoped access. Useful for the first deployment, for a second opinion on an architecture, and for surge capacity when a rollout is time-boxed.
Both, usually: The pattern that works is an external engineer leading the first deployment with your own people certified alongside them, then handing over. The credential is what makes that handover a transfer rather than a hope.
The reason a large deployment slows down is almost never the platform. It is the calendar of the three people who have done one before. A certified engineer inside your organisation removes that dependency entirely.
Everyone certified builds against the same Deployment Standard, so a Flow can be handed from an external partner to your own team without a rewrite, and a review means the same thing in both directions.
Two or three certified people inside a company is usually enough to make it self-sufficient for everything except architecture decisions, and to make the next department's rollout an internal project rather than a new procurement.
The assistant is on the platform at every step, which is what makes a two-day curriculum enough to reach production.
Drafted, then judged: The assistant drafts a Flow, an Automation or a tool definition from a description of the job, and explains what it did. The certified engineer's work is deciding whether that is right, which is the part the credential is about.
Testable out of the box: Simulated callers, generated edge cases and a suite you can re-run against every version. Nobody has to build a test harness first, which is the usual reason an enterprise rollout ships with none.
Traced by default: Every run, call, tool invocation and decision is on the record with the version that produced it. An auditor's question has an answer that does not depend on whether somebody kept notes.
Versioned and reversible: Versions, staged rollout and rollback are the platform's, not a convention a team agreed on. A new engineer inherits them on their first day instead of learning somebody else's discipline.
A traditional enterprise curriculum is long because most of it teaches the scaffolding: how to write the tests, how to wire the logging, how to keep the documentation honest. Here that scaffolding is the product, so the training can spend its time on judgement, which is the only thing an examination can usefully measure.
Enterprise packages include academy seats. Examination and the advanced credentials are charged separately, and authorised delivery partners are required to keep a minimum number of active certified professionals.
Sign in with your Kilevel account to reach your courses, examinations and credentials.
