GenAI for developers

What is AI?

• Artificial Intelligence (AI) = Computers performing tasks that normally need human intelligence
• Key abilities:
  • Learning & improving
  • Reasoning & decision-making
  • Understanding language
  • Recognizing patterns/images/sounds
  • Acting autonomously

What is AI NOT?

• Magic
• A crazy that will replace everything we do

How is it going?

• People continue to overestimate AI
• We don't have General Artificial Intelligence Yet
• AI is helping us write documents, generate memes, use energy, generate code
• Hallucinations - a bad word for

Why AI Won’t Replace Humans (Yet)

• Creativity & Imagination → AI recombines patterns, humans invent
• Common Sense & Context → struggles with nuance & ambiguity
• Emotions & Empathy → simulates, but doesn’t feel
• Ethics & Values → no built-in sense of right/wrong
• Real-World Adaptability → good in narrow tasks, poor in messy situations

Bottom line:
AI = fast & scalable
Humans = creative, ethical, adaptable
Together, they’re stronger.

How Does an LLM Work?

• Training: Learns patterns from massive text datasets
• Neural Networks: Billions of parameters adjust to capture language structure
• Tokenization: Breaks text into tokens (chunks of words/characters)
• Prediction: For each token, predicts the most likely next token
• Iteration: Repeats prediction → builds sentences, paragraphs, conversations
• Fine-tuning & Alignment: Trained further to follow instructions, stay safe, and be useful

In short:
LLMs don’t “know” like humans — they predict patterns of language based on data.

What is RAG?

• Retrieval: Searches external knowledge (docs, databases, web)
• Augmentation: Injects retrieved info into the prompt
• Generation: LLM uses both context + knowledge to answer

Why it matters:

• Reduces hallucinations
• Keeps answers up-to-date
• Combines AI reasoning with real data

In short:
RAG = Search + AI → More accurate, grounded answers

Preparing RAG

• Collect Knowledge → domain docs, PDFs, web pages, FAQs
• Chunking → split into small, retrievable pieces
• Embedding → convert chunks into vector representations
• Vector Database → store embeddings for fast semantic search

Steps of a RAG query

Retrieval → find the most relevant chunks at query time
↓
Augmentation → inject retrieved context into the prompt
↓
** LLM Generation** → LLM produces grounded Grounded Response

Solution on a Page and demo