// Slide 08: Evaluators + run_on_dataset -- off-line evaluation pipeline
// Code slide: custom evaluator + run_on_dataset.

const ds = require('./design-system');

function createSlide(pres, theme) {
  const slide = pres.addSlide();
  ds.helpers.slideBase(slide, pres, theme);

  ds.helpers.addHeader(slide, pres, theme, {
    eyebrow: 'STAGE 5: ECOSYSTEM',
    section: 'Evaluators',
    title: 'run_on_dataset + evaluator-ы',
    sectionNumber: 5,
    titleSize: 24,
  });

  ds.helpers.addCodeBlock(slide, pres, theme, {
    x: 0.5, y: 1.45, w: 9.0, h: 2.85,
    language: 'python',
    fontSize: 10,
    code: [
      'from langsmith import Client',
      'from langsmith.schemas import Example, Run',
      '',
      'client = Client()',
      '',
      '# 1. Target -- что прогоняем (chain, graph, agent, callable)',
      'def target(inputs: dict) -> dict:',
      '    return {"answer": my_chain.invoke(inputs["question"])}',
      '',
      '# 2. Evaluator -- scoring function',
      'def answer_match(run: Run, example: Example) -> dict:',
      '    score = 1.0 if example.outputs["answer"] in run.outputs["answer"] else 0.0',
      '    return {"key": "answer_match", "score": score}',
      '',
      '# 3. Прогон: target на датасете + scoring',
      'client.run_on_dataset(',
      '    dataset_name="rag-qa-eval-v1",',
      '    llm_or_chain_factory=target,',
      '    evaluators=[answer_match],',
      ')',
    ].join('\n'),
  });

  ds.helpers.addCallout(slide, pres, theme, {
    x: 0.5, y: 4.5, w: 9.0, h: 0.55,
    kind: 'success',
    text: 'Готовые evaluator-ы: llm_as_judge, exact_match, embedding_distance, cot_qa.',
  });

  ds.helpers.addSourceLine(slide, pres, theme, {
    source: 'docs.smith.langchain.com Evaluation > Evaluator types',
  });
  ds.helpers.addPageNumber(slide, pres, theme, 8);
}

module.exports = { createSlide };