Introduction to Generative AI & LLMs — 3-Day Workshop

3-day workshop: LLMs, prompt engineering, fine-tuning (LoRA), RAG, deployment.

Instructor: Dr. Yaé Ulrich Gaba Duration: 3 days (18 hours) Level: Intermediate Language: English

Overview

This workshop provides a practical, hands-on introduction to generative AI and Large Language Models (LLMs). Participants learn prompt engineering, fine-tuning, Retrieval-Augmented Generation (RAG), and deployment strategies. The workshop balances conceptual understanding with practical implementation using both commercial APIs (OpenAI) and open-source models.

Prerequisites

  • Basic Python programming (variables, functions, loops)
  • Familiarity with machine learning concepts (training, inference, evaluation)
  • An OpenAI API key (free tier available at platform.openai.com)
  • Laptop with Python 3.10+ installed

Learning Objectives

By the end of this workshop, participants will be able to:

  1. Understand the architecture and capabilities of modern LLMs
  2. Write effective prompts using structured prompt engineering techniques
  3. Build applications using the OpenAI API and LangChain
  4. Implement Retrieval-Augmented Generation (RAG) pipelines
  5. Fine-tune open-source models on custom datasets
  6. Evaluate LLM outputs and deploy applications responsibly

Software Requirements

  • Python 3.10+
  • Libraries: openai, langchain, transformers, sentence-transformers, chromadb, gradio
  • Optional: HuggingFace account, GPU access (Google Colab free tier works)

Day-by-Day Program

Day 1: Understanding LLMs & Prompt Engineering

Objectives: Understand how LLMs work and master prompt engineering techniques.

Time Topic
09:00–10:00 The Generative AI Landscape — From GPT to Claude to open-source: timeline, key models, capabilities and limitations, tokens and context windows
10:00–10:45 How LLMs Work — Transformer architecture (intuition, not math-heavy), pre-training, instruction tuning, RLHF, emergent capabilities
10:45–11:00 Break
11:00–12:30 Prompt Engineering Fundamentals — Zero-shot, few-shot, chain-of-thought, role prompting, system prompts, structured output (JSON mode), temperature and top-p
12:30–14:00 Lunch
14:00–15:30 Advanced Prompting — Prompt chaining, self-consistency, tree-of-thought, meta-prompting, prompt templates, handling long contexts
15:30–15:45 Break
15:45–17:00 OpenAI API — Authentication, chat completions, streaming, function calling / tool use, vision API, cost management

Lab 1: Build a prompt engineering toolkit: create a set of reusable prompt templates for common tasks (summarization, extraction, classification, code generation). Test each with different models and compare outputs.

Homework: Use the OpenAI API to build a script that summarizes academic papers from their abstracts.

Day 2: Building Applications — RAG & LangChain

Objectives: Build practical LLM-powered applications using RAG and orchestration frameworks.

Time Topic
09:00–09:30 Homework Review
09:30–10:30 Embeddings & Vector Search — What are embeddings, sentence-transformers, similarity search, vector databases (ChromaDB, FAISS), indexing strategies
10:30–10:45 Break
10:45–12:30 Retrieval-Augmented Generation — RAG architecture, document loading (PDF, web, CSV), chunking strategies, retrieval + generation pipeline, handling hallucinations
12:30–14:00 Lunch
14:00–15:30 LangChain Essentials — Chains, agents, tools, memory, output parsers. Building a multi-step reasoning agent
15:30–15:45 Break
15:45–17:00 Building a Q&A System — End-to-end: ingest documents, build vector index, create retrieval chain, add conversation memory, deploy with Gradio

Lab 2: Build a RAG-powered Q&A chatbot that answers questions about a collection of research papers (PDFs provided). The system should cite its sources and handle “I don’t know” gracefully.

Homework: Extend the chatbot to handle a new document collection relevant to your research/work.

Day 3: Fine-Tuning, Evaluation & Deployment

Objectives: Fine-tune models, evaluate outputs rigorously, and deploy responsibly.

Time Topic
09:00–09:30 Homework Review
09:30–10:30 Fine-Tuning Concepts — When to fine-tune vs. prompt vs. RAG, data preparation, instruction format, LoRA and parameter-efficient methods
10:30–10:45 Break
10:45–12:00 Hands-On Fine-Tuning — Fine-tuning a small open-source model (e.g., Mistral 7B / Llama) on a custom dataset using HuggingFace Transformers + PEFT. Using Google Colab for GPU access
12:00–12:30 OpenAI Fine-Tuning API — Preparing JSONL data, launching fine-tuning jobs, evaluating results, cost considerations
12:30–14:00 Lunch
14:00–15:00 Evaluation — Human evaluation, automated metrics (BLEU, ROUGE, BERTScore), LLM-as-judge, evaluation frameworks, red teaming basics
15:00–15:15 Break
15:15–16:00 Deployment & Ethics — API deployment, Gradio/Streamlit interfaces, rate limiting, content filtering, bias detection, responsible AI guidelines, African context considerations
16:00–17:00 Capstone & Wrap-Up — Present projects, discussion on LLM futures in research and industry, Q&A, certificates

Lab 3 (Capstone): Choose one project:

  • Research assistant: RAG chatbot for a specific research domain with citation support
  • Document analyzer: Automated extraction of key findings from a corpus of papers
  • Custom classifier: Fine-tuned model for domain-specific text classification (e.g., medical reports, legal documents)

Assessment

  • Daily labs (50%) — Working applications and code quality
  • Capstone project (30%) — End-to-end application presented on Day 3
  • Participation (20%) — Engagement and homework

Resources

Certificate

Participants who complete all labs and the capstone project receive a certificate of completion.