LLM Batch Pipeline

LLM Batch Pipeline#

The LLM Batch Pipeline is a tool for running large, auditable batches of LLM requests against tabular research data. It converts tabular data into structured, inspectable prompts, submits them to a provider, and standardizes the responses for analysis — separating preparation from execution so you can review prompts and cost estimates before anything is submitted.

Documentation and Source Code#

Maintained by Kellogg Research Support.

Key Features#

  • Immutable, inspectable prepared requests before submission

  • Cost estimation and provider limit validation

  • Record-level provenance tracking, with raw responses and retries stored separately

  • Source files remain unmodified

  • Support for per-record file attachments (PNG/PDF)

  • Retry management, provider comparisons, and batch cancellation with recovery

Supported Providers#

The pipeline supports OpenAI and Anthropic APIs, as well as self-hosted models served through Ollama or vLLM — including deployments on KLC. See Open Source LLMs on KLC and LLM API Usage for the underlying KLC workflows.