Month 5 — LLM, RAG and AI Agents

🎯 Goal of this month

By the end of the month you will be able to:

• Know the architecture and ecosystem of LLM (Large Language Model)
• Know how to work with OpenAI, Anthropic, Google AI APIs
• Apply prompt engineering techniques
• Build Vector DB and RAG (Retrieval Augmented Generation) pipelines
• Create AI Agents (tool use, function calling)
• Know how to do fine-tuning with LoRA/QLoRA
• Create chatbots for Uzbek-language documents

Weekly breakdown

Chapter order

1. LLM fundamentals — how GPT, Claude, Llama work
2. Prompt Engineering — writing good prompts
3. OpenAI and Anthropic API — practical work
4. LangChain and LlamaIndex — frameworks
5. Vector Databases — Qdrant, ChromaDB, pgvector
6. RAG Pipeline — full RAG implementation
7. AI Agents — tool use, multi-agent
8. Fine-tuning — LoRA, QLoRA, PEFT
9. Exercises

What can you do by the end of the month?

• Build a full chatbot with LLM API
• Build a RAG pipeline from 1000+ documents
• Multi-agent AI systems (CrewAI, LangGraph)
• Uzbek-language documentation bot
• Small domain-specific fine-tuning with LoRA
• Ship to production: streaming, caching, observability

Tip for Backend Dev

Working with LLM — 80% prompt engineering + 20% code. As a backend dev, your strengths are:

1. API integration — REST, streaming, retry logic
2. Schema design — structured output (Pydantic + JSON)
3. Caching and cost optimization — with Redis
4. Async/concurrent — async LLM calls
5. Observability — logging every LLM call

LLM API budget

For this month $10-30 is enough:

• OpenAI: GPT-4o-mini (very cheap — $0.15 per 1M tokens)
• Anthropic: Claude Haiku 4.5 (Sonnet 4.6 is also affordable)
• Google: Gemini 2.5 Flash (has a free tier)
• Groq: free (Llama, Mixtral models)
• OpenRouter: a single API for many models

**Recommendation:**top up $10 on OpenRouter — enough to try all models.

Start

Start with LLM fundamentals.