APIs

What is this?

An API (Application Programming Interface) lets your script or application talk directly to an LLM service. Instead of manually typing prompts in a chat window, you write code (e.g., Python, R) to send requests and receive responses.

When should you use it?

  • You need repeatable workflows
  • You want to process many documents automatically
  • You need integration with research scripts, notebooks, or internal tools
  • You want better logging, evaluation, and versioning
  • You need control over input/output formats and parameters

When should you NOT use it?

  • When your use case is occasional and exploratory only
  • When your team has no capacity for basic scripting support
  • When policy constraints prevent use of external API providers

How it works (simple explanation)

Your code sends structured input (prompt, parameters, optional context) to an API endpoint. The API endpoint is, in short, a URL that accepts such input and returns model output in a machine-readable format. This enables automation and consistent processing logic.

Concrete examples (tools/platforms)

  • Provider APIs (commercial or institutional). For example:
  • API client libraries in Python, R, or other languages. These libraries simplify the process of sending requests and handling responses. For example:
    • LangChain in Python for pipeline orchestration
    • ellmer in R for conversational and prompt workflows in data science contexts

Example workflow (step-by-step)

  1. Define a single research task, such as coding open-ended survey responses.
  2. Create a prompt template with clear output schema.
  3. Implement a script that reads rows from your dataset.
  4. Send each row to the API and parse the response.
  5. Store outputs with metadata (model, prompt version, timestamp).
  6. Evaluate quality on a validation subset and revise prompt/template.

Pros and cons

Pros Cons
Enables automation and scale Requires scripting and API key management
Supports reproducible pipelines Cost management becomes important at volume
Easier integration with existing tools Governance checks may be more complex

Learning resources