AI Fundamentals

Key Concepts

LLMs (Large Language Models)

Neural networks trained on massive text datasets to predict and generate text. Core of ChatGPT, Claude, Gemini.

What they’re good at:

• Text generation and summarization
• Code generation
• Question answering
• Translation and transformation

What they struggle with:

• Precise arithmetic
• Real-time information
• Deterministic outputs
• Long-term memory (without tooling)

Tokens

LLMs don’t read words — they read tokens (roughly 3/4 of a word).

Model	Context window
Claude Sonnet 4.6	200k tokens
GPT-4o	128k tokens
Gemini 1.5 Pro	1M tokens

Embeddings

Convert text into vectors (arrays of numbers) that capture semantic meaning. Used for:

• Semantic search
• Similarity matching
• RAG (Retrieval Augmented Generation)

RAG (Retrieval Augmented Generation)

User query
    ↓
Embed query → search vector DB → retrieve relevant chunks
    ↓
Inject chunks into LLM prompt
    ↓
LLM generates answer grounded in your data

Prompt Engineering

System vs User prompts

messages = [
    {"role": "system", "content": "You are an Odoo expert."},
    {"role": "user", "content": "How do I create a custom module?"}
]

Few-shot prompting

Give examples in the prompt to guide output format:

Extract invoice data as JSON.

Example:
Input: "Invoice #123 from Acme Corp, $500 due Jan 15"
Output: {"number": "123", "vendor": "Acme Corp", "amount": 500, "due": "Jan 15"}

Now extract:
Input: "Bill #456 from TechCo, $1200 due Feb 28"

Claude API Quick Start

import anthropic

client = anthropic.Anthropic()

message = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    messages=[
        {"role": "user", "content": "Explain Odoo's ORM in 3 sentences."}
    ]
)

print(message.content[0].text)

AI in Functional Consulting

Practical applications:

• Requirements extraction — summarize meeting notes into user stories
• Gap analysis — compare AS-IS vs TO-BE process documents
• Test case generation — generate UAT scripts from requirements
• Documentation — auto-generate functional specs from data models