LLM Zoomcamp 2026
Personal solutions and capstone for LLM Zoomcamp 2026 — DataTalksClub.
Repo: llm-zoomcamp-2026
End-to-end LLM engineering course by DataTalksClub, covering the full stack for production-grade RAG systems and AI agents — from retrieval and embeddings through orchestration, evaluation, and monitoring. Each module is a Jupyter notebook with a concrete homework, building toward a capstone that integrates the full stack.
Modules
| # | Topic | Key concepts | Status |
|---|---|---|---|
| 1 | Agentic RAG | RAG pipeline, minsearch, chunking, agentic loop with function calling | Done |
| 2 | Vector Search | Embeddings, ONNX Runtime, cosine similarity, hybrid search with RRF | Done |
| 3 | Orchestration | Kestra workflows, context engineering, RAG grounding, token usage observability | Done |
| 4 | Evaluation | Hit Rate, MRR, ground-truth generation, text/vector/hybrid retrieval comparison | Done |
| 5 | Monitoring | OpenTelemetry tracing, spans, token/cost attributes, custom SQLite span exporter | Done |
| — | Workshop: dlt + Logfire | Pydantic AI agent tracing, dlt data pipelines, DuckDB, nested JSON normalization | Done |
| — | Capstone | Full RAG/agent system — ingestion, eval, UI, monitoring | Done — see write-up |
Stack
OpenAI GPT-4o-mini / GPT-5.4-mini, Google Gemini · ONNX Runtime (all-MiniLM-L6-v2) · minsearch, ElasticSearch, pgvector · Pydantic AI, LangGraph, CrewAI · Kestra · OpenTelemetry, Pydantic Logfire · dlt, DuckDB · Streamlit, PostgreSQL, Grafana · Docker Compose, uv