Oy 5 — Mashqlar to'plami

🟢 Easy

LLM Fundamentals

1. tiktoken bilan inglizcha va o'zbekcha matnda token solishtirish.
2. 5 ta modelni (GPT-4o-mini, Claude Haiku, Gemini Flash, Llama 3.1, Mistral) bir xil savol bilan.
3. Temperature 0, 0.5, 1.5 — javob farqlarini ko'rish.

Prompt Engineering

1. Zero-shot, few-shot, CoT — bir xil masala uchun.
2. Instructor bilan structured Pydantic output.
3. JSON output uchun prompt + validation.

APIs

1. OpenAI streaming chat.
2. Anthropic prompt caching.
3. Function calling — 3 ta tool.

Vector DB

1. ChromaDB'da 100 ta hujjat.
2. Qdrant Docker setup.
3. pgvector Postgres extension.

RAG

1. Naive RAG — 10 hujjat, query.
2. Citation — [Source N] format.
3. Chunking strategiyalari solishtirish.

Agents

1. Pydantic AI agent + 3 tool.
2. CrewAI hello world.
3. LangGraph oddiy workflow.

Fine-tuning

1. Pretrained Llama 1B yuklash.
2. 50 ta sintetik dataset (GPT bilan).
3. LoRA config sintaksis tushunish.

🟡 Medium

Real loyihalar

1. Multi-turn chatbot: history saqlash, context window manage.
2. RAG over Wikipedia: 100 ta o'zbek Wikipedia maqolasi.
3. PDF Q&A bot: PyPDF + Qdrant + Streamlit.
4. Code review agent: GitHub PR diff → suggestions.
5. Email summarizer: 50 ta email → daily digest.

Advanced techniques

1. Multi-query RAG: query expansion bilan.
2. HyDE: hypothetical embeddings.
3. Hybrid search: dense + BM25.
4. Reranking: cross-encoder bilan.
5. Multi-agent: CrewAI 3 agentli tizim.

Fine-tuning

1. TinyLlama: o'zbek instruction dataset bilan QLoRA (Colab).
2. OpenAI fine-tuning: customer support classifier ($1 budget).
3. Sintetik data: GPT-4 yordamida 500+ training pairs.

🔴 Hard (Production)

1. Documentation Q&A Bot

Talab:

• 100+ ta hujjat (PDF, markdown, websites) ingestion
• Qdrant + FastAPI + Celery
• Multi-query + reranking
• Citation va source links
• Streamlit UI
• Langfuse observability
• Cost tracking per user

2. AI Customer Support Agent

Talab:

• Telegram bot (aiogram)
• Multi-turn conversation
• Tools: FAQ search, order lookup, refund process, escalate to human
• LangGraph workflow
• Postgres memory
• Sentiment-based routing
• Admin dashboard

3. RAG Evaluation Framework

Talab:

• Test set yaratish (100+ Q&A pairs)
• RAGAS bilan automated evaluation
• A/B testing framework
• Continuous improvement loop
• Grafana dashboard

4. Domain-specific Fine-tuning Pipeline

Talab:

• Data collection + cleaning
• Synthetic data augmentation
• QLoRA fine-tuning (Llama 3.1 8B)
• vLLM serving
• Benchmark (vs base model)
• Production rollout strategy

Mini-loyihalar

Mini-loyiha 1: Voice-to-Text Meeting Assistant

• Whisper (audio transcription)
• LLM summarization
• Action items extraction
• Slack integration

Mini-loyiha 2: Code Review Bot

• GitHub webhook
• Diff parsing
• LLM analysis (security, performance)
• Inline PR comments

Mini-loyiha 3: Personal Knowledge Base

• Notion + Obsidian export
• Vector DB ingestion
• "Second brain" chatbot
• Smart search

Mini-loyiha 4: O'zbek Tilidagi Hukumat Hujjatlari Chatbot

• lex.uz, data.gov.uz scraping
• Multi-language (uz/ru)
• Citation
• Legal disclaimer

Quiz

LLM

1. Token, context window, temperature, top_p — har birini tushuntiring.
2. Pretraining, SFT, RLHF — qanday navbat?
3. Hallucination nima va qanday kamaytirish?
4. Proprietary vs Open Source LLM — tanlov mezonlari?
5. Prompt caching qanday ishlaydi?

Prompt Engineering

1. Zero-shot, few-shot, CoT qachon qaysi?
2. Structured output (JSON) uchun pattern'lar?
3. Prompt injection — xavf va himoya?
4. Self-consistency texnikasi?
5. ReAct pattern intuition?

RAG

1. RAG vs Fine-tuning farqi?
2. Chunking strategiyalari trade-off?
3. HNSW algoritm qanday ishlaydi?
4. Hybrid search nima?
5. Cross-encoder reranking nima uchun yaxshilanish keltiradi?

Agents

1. Agent va LLM call farqi?
2. ReAct pattern — Thought/Action/Observation?
3. Multi-agent qachon kerak?
4. MCP (Model Context Protocol) nima?
5. Agent xavfsizligi — sandbox patternlar?

Fine-tuning

1. LoRA mathematik intuition?
2. QLoRA — nima uchun 4-bit?
3. Sintetik data generation strategiyalari?
4. RAG vs Fine-tuning — qachon birinchisini, qachon ikkinchisini?
5. vLLM nima uchun production'da tez?

✅ Oy 5 oxiri checklist

• LLM API'larni (OpenAI, Anthropic) ishlataman
• Prompt engineering texnikalarini bilaman
• Structured output (Instructor, Pydantic AI)
• Vector DB (kamida 2 ta) bilan tanish
• To'liq RAG pipeline yaratdim
• AI Agent (tool use) yozdim
• LoRA bilan kichik fine-tuning sinab ko'rdim
• Production'ga olib chiqdim (FastAPI + Docker)
• Langfuse / observability
• Capstone loyiha (chatbot/RAG)
• LinkedIn'ga post

Tabriklayman! Oy 6 — MLOps va Production — sizning asosiy maqsadingiz uchun eng muhim oy.