romanoremo.it
All courses
AI & Automation

LLMs & Advanced Prompt Engineering

Anyone who has used AI assistants for a while reaches the same point: the results are decent but inconsistent, and cannot be placed inside a process you would trust. This course addresses exactly that gap.

We start from a conceptual understanding of how language models work, enough to see why certain techniques help, then move to those techniques: chain of thought, few-shot, tree of thought, structuring output as JSON, and systematic evaluation of responses.

The second half covers RAG and integration: building a company knowledge base the model can consult, and connecting an LLM to real workflows via API. Along with Building AI Agents, it is the most technical course in the catalogue.

Indicative duration
20 hours
Level
Advanced
Prerequisites
Daily use of AI assistants. Basic familiarity with APIs and JSON is useful, though not as a developer.

Who it is for

  • People already using AI assistants daily who want to go further
  • Technical teams integrating LLMs into their processes
  • Automation and knowledge management project leads
  • Consultants building solutions for clients

What you will be able to do

  • Understand conceptually how a language model reasons
  • Apply advanced prompting techniques systematically
  • Design a RAG system over company documents
  • Produce reliable structured output in JSON
  • Evaluate responses and reduce hallucinations
  • Integrate an LLM via API into an existing workflow

Programme

  1. 01

    Large Language Model architecture (conceptual)

    Tokens, context, temperature and training cutoffs. No mathematics, but enough to predict how the model will behave.

  2. 02

    Advanced techniques: chain of thought, few-shot, tree of thought

    The strategies that measurably improve results on complex tasks, and the cases where they make things worse.

  3. 03

    RAG: building company knowledge bases

    Retrieval Augmented Generation: giving the model access to company documents. Chunking, embeddings, retrieval and answer composition.

  4. 04

    Structured output: JSON, markdown and structured data

    Getting responses in a format another system can consume, with validation and handling for when the format breaks.

  5. 05

    Evaluation, fact-checking and reducing hallucinations

    Building a test set, measuring quality over time, and adding automated checks before output reaches a user.

  6. 06

    LLM integration via API in real workflows

    API calls, managing cost and latency, caching and fallbacks. How a demo becomes something that runs every day.

Tailored programmes

No course is a fixed package. Duration, angle and depth are agreed around the objectives of the people taking part, and topics from different courses can be combined into a single programme. I work with corporate groups and with one person alone: tell me what you need and we will build the programme.

Request information

Frequently asked questions

Do I need to be able to code?

Not strictly, but the RAG and API integration modules give considerably more to people with some coding familiarity.

How does it differ from AI for Business?

AI for Business is introductory and non-technical. This is the next step: it systematises usage and addresses integration into processes.

Do we build a working RAG?

Yes, it is the central exercise. In corporate programmes we use real company documentation.

Which models are used?

We work with the main available models and compare them. The techniques taught are provider independent.

Related courses