NLP & Large Language Models

Module 04 · Days 22–28

From "what is a token" to a deployed RAG chatbot in 7 days. Use APIs from day one, understand internals by day 3, ship something real by day 7.

D22

NLP basics: tokenisation, embeddings & vocabulary

How text becomes numbers. BPE, word embeddings, and why context matters for meaning.

D23

Prompt engineering: system prompts, CoT & structured outputs

Writing prompts that reliably produce the output you want. Chain-of-thought, few-shot, JSON mode.

D24

RAG: retrieval-augmented generation

Ground LLM answers in your own data. Embeddings, vector search, chunking — the number one enterprise LLM pattern.

D25

Fine-tuning LLMs: LoRA & PEFT on free GPUs

Adapt a foundation model to your domain. LoRA on a free Colab GPU. When to fine-tune vs prompt-engineer.

D26

LLM evaluation: benchmarks, hallucinations & safety

Measuring model quality beyond vibes. MMLU, TruthfulQA, LLM-as-judge, and a custom eval suite.

D27

LLM APIs in production: cost, latency & model routing

Token costs, rate limits, routing between cheap and expensive models, prompt caching. Real engineering.

🏗

🏗️ PROJECT: RAG Chatbot with Citations + RAGAS Evaluation

Domain RAG chatbot over real documents. Streaming UI, source citations, and full RAGAS evaluation report.