LLM tutorials: integration, task workflows, evaluation and operations

Choose a guide for the problem you are solving. For a new integration, start with a minimal request, then check protocol behavior and error handling.

Connect or migrate an application

• Send your first LLM request — Discover a compatible model and inspect the response and usage.
• Choose OpenAI, Claude or Gemini protocol — Match authentication, request envelopes and response parsing to the route.
• Migrate an OpenAI Node.js SDK client — Check base URL, model IDs and supported features before moving an existing client.
• Prepare Gemini video-analysis requests — Check bridge input handling and preserve the native response context.

Build a task workflow

• Review code with GPT-6.1 Sol — Ask for findings grounded in the supplied code, then verify them.
• Extract evidence with Claude Fable 5.1 — Keep source quotations and distinguish missing facts from inferred answers.
• Create a Gemini video shot log — Draft a log of an existing clip and verify timestamps against the source.
• Choose full context, chunks or retrieval — Decide how to supply long documents without losing cross-section evidence.
• Classify tickets with abstention — Validate labels and escalate ambiguous cases instead of forcing an answer.

Compare results and maintain prompts

• Design a task-based evaluation — Define a dataset, rubric and comparable conditions before measuring models.
• Compare GPT, Claude and Gemini support drafts — Apply a shared support brief and review groundedness, completeness and cost.
• Check prompt-version regressions — Use paired fixtures and offline validators to identify newly failing behavior.

Handle streaming, failures and cost

• Handle streaming responses — Separate partial output from completion and handle interrupted streams.
• Diagnose errors and bounded retries — Distinguish contract errors, rate limits and uncertain outcomes before retrying.
• Reconcile cost and usage — Separate estimates from account pricing and actual usage receipts.

Model contracts and related guides

Use the Common API for discovery, model documentation for model-specific parameters and the complete API reference for endpoints. Analyzing an existing video differs from generating a new one; use the image and video tutorials when you need to generate assets.

These guides provide methods and acceptance criteria, not performance rankings or live pricing promises. Confirm account access, the current model contract and budget before running generation examples.