VEL ORENDISVel Orendis

Build an evaluated RAG assistant

A grounded assistant with measured retrieval, judged answers and live monitoring.

24 lessons, about 4.2 h. Anything you already know, you can prove and skip.

0 of 24 lessonsquests defended, 1 open now

  1. The request path and where time goesFoundations, 10 min
    Opens later
  2. Tables, keys and modellingDatabases, 10 min
    Opens later
  3. Observability: logs, metrics, tracesReliability, 12 min
    Opens later
  4. What the Model Cannot KnowWorking with LLMs, 8 min
    Open
  5. Calling a model: messages, tokens and costWorking with LLMs, 10 min
    Opens later
  6. Prompts that ground answersWorking with LLMs, 10 min
    Opens later
  7. Structured output and validationWorking with LLMs, 10 min
    Opens later
  8. The RAG pipelineSearch and RAG, 10 min
    Opens later
  9. Lexical search and BM25Search and RAG, 12 min
    Opens later
  10. Embeddings and similaritySearch and RAG, 12 min
    Opens later
  11. Vector search, exact and approximateSearch and RAG, 12 min
    Opens later
  12. Chunking long documentsSearch and RAG, 12 min
    Opens later
  13. Ground truth for searchRetrieval quality, 10 min
    Opens later
  14. Hit rate, MRR and other metricsRetrieval quality, 12 min
    Opens later
  15. Hybrid search and rank fusionRetrieval quality, 12 min
    Opens later
  16. Query rewriting, thresholds and the honest no-answerRetrieval quality, 10 min
    Opens later
  17. Offline and online evaluationEvaluation and monitoring, 8 min
    Opens later
  18. LLM as a judgeEvaluation and monitoring, 12 min
    Opens later
  19. Logging every LLM callEvaluation and monitoring, 10 min
    Opens later
  20. Feedback and online judgesEvaluation and monitoring, 10 min
    Opens later
  21. Dashboards and traces for LLM appsEvaluation and monitoring, 12 min
    Opens later
  22. Testing strategyShipping, 10 min
    Opens later
  23. Configuration, secrets and containersShipping, 10 min
    Opens later
  24. From notebook to serviceShipping, 10 min
    Opens later