AI Specialization Program

WEEK 5 :

NLP, Transformers & Generative AI Engineering

Theme: Building Modern LLM-Powered Systems

Week 5 focuses on:

Understanding and engineering systems around Large Language Models.

Not just using ChatGPT —

building controlled, production-aware GenAI systems.

Week 5 —

6 Lesson Structure

Lesson 1:

NLP Foundations

Focus:

Text as data

Tokenization

Embeddings intuition

Text preprocessing

Traditional NLP vs modern NLP

Outcome: Students understand how machines process language numerically.

Lesson 2:

Transformers & LLM Architecture

Focus:

Attention mechanism intuition

Transformer architecture overview

Encoder vs Decoder models

Why LLMs scale

Context windows

Context Engineering fundamentals

Context quality and grounding

Prompt context vs retrieved context

Context window limitations

Long-context challenges.

Outcome: Students understand how modern LLMs work internally and use context and why context engineering is critical for performance.

 Lesson 3:

Hugging Face Ecosystem

Focus:

Hugging Face Datasets

Transforrs library

Loading pre-trained models

Running inference

Model cards & documentation

Outcome: Students can use pre-trained transformer models properly.

Lesson 4:

Prompt Engineering & LLM Evaluation

Focus:

Prompt structuring techniques

Role prompting

Few-shot prompting

Chain-of-thought

Hallucination awareness

Basic LLM evaluation methods

JSON outputs

Schema-based responses

Reliable machine-readable outputs

LLM Reliability Testing

Consistency testing

Prompt robustness

Edge-case evaluation

Reliability measurement

Outcome:

Students can design reliable, structured, and testable LLM interactions

 Lesson 5:

RAG Systems & Simple Agents

Focus:

What is Retrieval-Augmented Generation (RAG)?

Vector databases (FAISS / Chroma concept)

Embedding search intuition

Simple agent concepts (ReAct / tool-calling idea)

Controlled generation systems

Function Calling:

What function calling is

Why modern LLMs call tools

Structured action execution

Tool Use:

Search tools

Databases

APIs

Model Context Protocol (MCP) — Standardizing How Agents Connect to Tools

Calculators

External systems

Modern Agent Workflows:

Reasoning

Planning

Tool execution

Response generation

Outcome:

Students can design RAG systems, tool-enabled LLM systems, and basic AI agent workflows.

 Lesson 6: Fine-Tuning & GenAI Integration Project

Focus:

Concept-level LoRA / QLoRA

When to fine-tune vs prompt

Lightweight adaptation methods

Mini project: Build a small RAG or GenAI system

Documentation & evaluation

Agent Evaluation:

Task completion evaluation

Tool-use evaluation

Response quality evaluation

Reliability measurement

Modern GenAI System Design:

Students learn to think in terms of:

Input → Context → Retrieval → Reasoning → Tool Use → Validation → Output

instead of:

Prompt → Response

Outcome:

Students can evaluate and design modern GenAI systems beyond basic prompting. They can build their first structured GenAI system.

Week 5 Final Outcome

By the end of Week 5, students can:

✔ Explain how NLP systems process text

✔ Understand Transformer & LLM architecture

✔ Use Hugging Face models properly

✔ Design effective prompts

✔ Evaluate LLM outputs responsibly

✔ Build basic RAG systems

✔ Understand fine-tuning concepts

✔ Engineer structured GenAI workflows

Week 5 produces:

LLM-capable AI engineers. responsibly.

Powered by Soft AI Africa | Training the Next Generation of AI Leaders in Africa.