Skip to content
Giallo

AI for training providers: agents for organizations that design, deliver and report on courses.

Where does AI help a training provider?

AI for training providers helps before and after the classroom: AI agents that set up course design around funding calls, prepare materials and self-assessment quizzes, check registers and documents for funder reporting and answer learners. Giallo Studio designs them to measure with Claude, GPT and Gemini; assessments, certificates and signatures stay with trainers and managers.

Frequently asked questions, straight answers.

Yes, as a first draft to review. The agent proposes structure, objectives, materials and quizzes based on the funding call and the trainers' content; teaching choices and content review stay with the course designer and the trainer.

No, not in our projects. Exams, admissions and certificates stay with teachers and committees; the agent works on project proposals, reporting and communications with learners.

Yes, through ongoing document checks. It compares registers, hours and participant documents against the call's rules and flags discrepancies while the course is still open; the report is signed by the organization's manager.

With Moodle, yes, through its web services. For other platforms it depends on the APIs or exports available: we check this at the start and, if they're missing, the agent works on the reports the platform already produces.

It depends on the task. Claude Opus for reading long funding calls, Gemini for recorded lessons and scans, Claude Haiku for learners' organizational questions, GPT for extracting data into rigid schemas: the choice is confirmed during testing.

No. Here we're talking about organizations that design and deliver courses for others; anyone looking into AI for training their own staff will find a dedicated guide on our blog, with different needs and tools.

Yes, and it's often the simplest starting point. Without the constraints of a funding call, the agent works only on materials, quizzes and e-learning: anyone evaluating AI tools to create catalog training courses can try it on their next upcoming course.

Sources of answers: Does it work with Moodle or the e-learning platform we use?

The AI agents we design, one by one.

The sign-in sheet of the class register: the registers and reporting agent compares each time slot in the approved calendar with the trainer's initials on the scanned register. Complete days get a check mark; on Wednesday one set of initials is missing and the discrepancy goes to the course office, without the agent touching the register.

Funding call design agent

Reads the call and its annexes, extracts the constraints and proposes a first training project structure that the course designer completes.

When a regional funding call, a fund's call for proposals or a request from a client company comes out.

Connects to: Regional calls and interprofessional training fund calls · Project application form · Archive of the organization's projects

  1. Reads the call, the technical annexes and the application form and extracts target participants, requirements, and the constraints on duration, delivery methods and costs stated in the text.
  2. Links the project to the reference profiles and skills the course designer specifies.
  3. Proposes modules, learning objectives and assessment methods consistent with those constraints.
  4. Flags every point in the call that isn't reflected in the draft.
  5. Delivers the draft in the application form's format, ready for the course designer's review.
Claude Opus
Funding calls are long and the constraints sit in different places: we need a model that reasons over the whole text and doesn't lose a requirement between one annex and the next.
GPT
We compare it with Claude Opus during testing on closed calls, to choose the model whose drafts best respect the constraints.

Who approves. The project is signed by the head of course design: the agent doesn't submit applications and doesn't declare the organization's requirements.

Materials and e-learning agent

Turns the trainer's materials into summaries, glossaries, transcripts and self-assessment quizzes for the e-learning platform.

When a trainer hands in slides, handouts or a lesson recording.

Connects to: Moodle or the organization's e-learning platform · Lesson recordings · Trainers' materials

  1. Receives the trainer's materials and, for recorded lessons, prepares the transcript.
  2. Writes a summary for each module and a glossary of the technical terms used.
  3. Proposes self-assessment questions with the correct answer and a pointer to the relevant point in the lesson.
  4. Queues everything for the trainer, who corrects and approves it before it's uploaded to the platform.
Gemini
It works well on long recordings and on slides full of images and tables: we choose it when the source material is a video lesson.
Claude Sonnet
It writes clear summaries and questions while staying faithful to the trainer's material, without adding content the trainer didn't cover.

Who approves. Every piece of content published on the platform is approved by the trainer, who remains the author of the course.

Registers and reporting agent

Checks documents and attendance as they arrive and prepares the reporting file according to the funding call's rules.

With every signed register, timesheet or participant document that enters the course file.

Connects to: Course management software · Scanned class registers · E-learning platform reports

  1. Reads scanned class registers, sign-in sheets and tracking reports from the e-learning platform.
  2. Compares attendance, teaching hours and the approved calendar, and flags anything that doesn't add up.
  3. Checks that each participant has the documents required by the call.
  4. Prepares a weekly list of discrepancies for the course office.
  5. Once the course is over, it assembles the file in the required order, ready for the manager's check.
Gemini Flash
It reads large volumes of scanned registers and sign-in sheets quickly and cheaply, which is the most repetitive work in the whole file.
GPT
It returns attendance and hours in a rigid schema that the connector compares with the calendar, flagging every difference.
Codex
It writes and maintains the connectors to the management software and the e-learning platform through APIs, such as Moodle's web services, or exports; it doesn't read learners' data.

Who approves. Funder reporting is signed by the organization's manager: the agent flags discrepancies and prepares the file, but doesn't correct registers or change attendance.Source

Tutor and learner support agent

Answers learners about logistics and the platform and gathers content questions for the tutor.

Throughout the course, on the platform, by email or on WhatsApp.

Connects to: E-learning platform · WhatsApp Business Platform · Course calendar

  1. Answers questions about the calendar, locations, platform access, attendance rules and certificate timing using the course information sheets.
  2. For content questions it points to the place in the materials where the answer can be found, when there is one.
  3. Groups open questions by topic and passes them to the tutor with context.
  4. Alerts the course office when a learner is at risk of not reaching the minimum attendance set by the funding call.
Claude Haiku
Organizational questions are many, short and similar to each other: a fast, low-cost model is enough and leaves room for the real cases.
Claude Sonnet
When a question touches on content, it answers only with what's in the approved materials and passes everything else to the tutor.

Who approves. The coordinator approves the course information sheets; exceptions on attendance, make-up sessions and certificates are always decided by a person.

Your next course, your first agent

Tell us how you work today across funding calls, trainers' materials and reporting: we'll suggest the first agent to try on a course about to start.

Prefer the full form? Go to the contact page.

No spam or sales pitches. Response within 1 business day.

We use this data to contact you and, if we work together, to manage the project. We store it in Giallo's internal management system, on servers in the European Union, where only our team can see it; by email we only receive a notice, without your data. If you do not become a client, we delete it 3 months after our last contact. Details and your rights: privacy policy.

Do you prefer to write or call us?

Where time gets lost today.

These are the points we hear about most often. Each one is a possible first agent.

The funding call is out and the project has to be written in a few days

The course designer reads the call, the technical annexes and the reference occupational profiles, then puts together modules, objectives and hours, fitting in constraints scattered across dozens of pages.

An agent extracts the call's constraints into a summary sheet and proposes a first project structure consistent with them; the course designer works on the choices, not on the research.

Trainers' materials arrive in every format

Slides, handouts and lesson recordings arrive differently from each trainer, and the coordinator adapts them by hand for the e-learning platform, adding quizzes and summaries.

The agent prepares summaries, glossaries and self-assessment quizzes from the trainer's materials, and the trainer reviews them before they're published on the platform.

Reporting turns out to be incomplete at the end

Signed registers, trainer timesheets, e-learning tracking data and participant documents are all checked together once the course is over, when a missing signature costs days of searching.

The agent checks registers and documents as they come in and immediately flags missing signatures, hours that don't add up and missing attachments, while the course is still running.

Learners write at all hours

They ask about the lesson's location and time, how to access the platform, when the certificate will arrive, and the tutor answers in the evening between one course and the next.

Organizational questions get an immediate answer from the course calendar and rules, and questions about content reach the tutor already grouped by topic.

Which AI model we use, and for which task.

We don't tie you to a single provider: we work with Claude by Anthropic, GPT and Codex by OpenAI, and Gemini by Google, and we choose the version during testing, on your own cases.

Claude Opus

  • Funding call design agent. Funding calls are long and the constraints sit in different places: we need a model that reasons over the whole text and doesn't lose a requirement between one annex and the next.

GPT

  • Funding call design agent. We compare it with Claude Opus during testing on closed calls, to choose the model whose drafts best respect the constraints.
  • Registers and reporting agent. It returns attendance and hours in a rigid schema that the connector compares with the calendar, flagging every difference.

Gemini

  • Materials and e-learning agent. It works well on long recordings and on slides full of images and tables: we choose it when the source material is a video lesson.

Claude Sonnet

  • Materials and e-learning agent. It writes clear summaries and questions while staying faithful to the trainer's material, without adding content the trainer didn't cover.
  • Tutor and learner support agent. When a question touches on content, it answers only with what's in the approved materials and passes everything else to the tutor.

Gemini Flash

  • Registers and reporting agent. It reads large volumes of scanned registers and sign-in sheets quickly and cheaply, which is the most repetitive work in the whole file.

Codex

  • Registers and reporting agent. It writes and maintains the connectors to the management software and the e-learning platform through APIs, such as Moodle's web services, or exports; it doesn't read learners' data.

Claude Haiku

  • Tutor and learner support agent. Organizational questions are many, short and similar to each other: a fast, low-cost model is enough and leaves room for the real cases.

How we choose between models: the best AI model for agents.

How we work with you.

  1. A course from funding call to certificate

    Together with the organization we retrace a course that's already finished, from design to reporting, to see where time and documents got lost.

  2. The call's rules, the agent's rules

    We turn the call's constraints and the organization's procedures into written instructions, with a list of what the agent never decides.

  3. Connection to the platform and management software

    We check what the e-learning platform and course management software offer and build connectors with access limited to the courses involved.

  4. Testing on a course about to start

    We trial the agents on a real course alongside the course office's usual work, compare the results and only then remove the double check where it makes sense.

AI agents for training providers

Which stage of the course weighs most on your organization?

Tell us how you work today across funding calls, trainers' materials and reporting: we'll suggest the first agent to try on a course about to start.