{"slug":"evidence-based-llm-development","title":"Evidence-Based LLM Development","published":true,
 "description":"Building LLM applications that can be measured: model APIs, retrieval components, and the evaluation and observability layer that turns prompt tinkering into something closer to engineering.",
 "items":[
  {"tool":"claude-developer-platform","blurb":"Anthropic's API and developer tooling."},
  {"tool":"openai-platform","blurb":"OpenAI's API and developer platform."},
  {"tool":"openrouter","blurb":"Many providers behind one API, useful for model comparisons."},
  {"tool":"sentence-transformers","blurb":"Embedding and reranking models you can run yourself."},
  {"tool":"voyage-ai","blurb":"Hosted embeddings and rerankers."},
  {"tool":"pgvector","blurb":"Vector similarity inside PostgreSQL, often enough instead of a separate store."},
  {"tool":"qdrant","blurb":"Dedicated vector database when filtering and scale demand it."},
  {"tool":"ragas","blurb":"Metrics for retrieval-augmented generation pipelines."},
  {"tool":"promptfoo","blurb":"Declarative test suites for prompts, models and agents."},
  {"tool":"langfuse","blurb":"Tracing, datasets and evaluations, self-hostable."},
  {"tool":"braintrust","blurb":"Hosted evaluation and logging platform."},
  {"tool":"dspy","blurb":"Programmatic prompt optimization against a metric."},
  {"tool":"instructor","blurb":"Structured, validated outputs from any model."}
 ]}