HarmonyFidelisHarmonyFidelis
Login
AI foundationsHow LLMs workLLM skillsEvaluationGlossaryBuilder

Aura · AI learning path

Understand AI. Build with LLMs. Verify what works.

A progressive, practical map of artificial intelligence—from machine learning and deep learning to tokens, Transformers, RAG, tools, agents, evaluation and safety. No magic words, no hype: concepts, decisions and checks you can reuse.

Start with the fundamentalsBuild an AI instruction
  • Free access
  • No account
  • Primary sources

01 · Aura · AI learning path

Six sections, one coherent path

Follow the order if you are starting out, or open the section that matches the problem you need to solve.

  1. 01 · 20 min

    AI, ML and deep learning

    Separate systems, models and products; understand how learning differs from fixed rules.

    Choose the right family of approach before choosing a model.

    Open section
  2. 02 · 25 min

    How LLMs work

    Tokens, embeddings, attention, Transformers, training, alignment and inference.

    Reason about capabilities and limits without treating an LLM as a database.

    Open section
  3. 03 · 30 min

    LLM engineering skills

    Prompt contracts, structured outputs, RAG, tools, MCP, agents and adaptation.

    Match each technique to the failure it can actually address.

    Open section
  4. 04 · 25 min

    Evaluation and safety

    Baselines, representative tests, factuality, injection, privacy, cost and rollout.

    Decide with measured evidence and limit the system’s blast radius.

    Open section
  5. 05 · Reference

    AI glossary

    Search 46 essential terms, from algorithm and gradient to RAG and quantization.

    Decode technical discussions and documentation quickly.

    Open section
  6. 06 · Tool

    Instruction builder

    Turn an objective, context and success criteria into a portable AI instruction.

    Produce a clear prompt you can inspect, edit and reuse.

    Open section

02 · Aura · AI learning path

The working map

Most useful LLM systems are pipelines. Quality depends on the weakest interface, not only on the model.

  1. 01

    Frame

    Define the user need, risk and observable success.

  2. 02

    Ground

    Provide relevant context, data or retrieval.

  3. 03

    Generate

    Use the model with explicit constraints and format.

  4. 04

    Act

    Authorize tools narrowly and validate every boundary.

  5. 05

    Evaluate

    Measure quality, safety, cost and failure recovery.

03 · Aura · AI learning path

Choose a path by objective

Core principle: a fluent answer is a proposal, not proof. Keep important claims traceable and important actions bounded.

I want to understand

Fundamentals → LLMs → Glossary

Build an accurate mental model and a durable vocabulary.

I want to build

LLMs → Skills → Evaluation

Design a useful system and test it against real cases.

I want to use AI better

Fundamentals → Builder → Evaluation

Give clearer instructions and review outputs critically.