LLM for Efficient Translation and Transformation into Uniform Clinical Encoding

Lettuce is an application for medical researchers that matches the source terms supplied by the user to concepts in the Observational Health Data Sciences and Informatics (OMOP) standardised vocabularies

The application can be used as an HTTP API, a CLI, or run with a graphical user interface (GUI).

Overview

Lettuce uses vector search, a large language model, and text search features to help find the matching concept for a source term. A full pipeline uses the vector search results for retrieval-augmented generation, then the answer provided by an LLM to run text search against a configured OMOP-CDM database. Users can use the full pipeline, or only components of it, depending on requirements.

Installation

To use Lettuce, follow the quickstart

Connecting to a database

Lettuce works by querying a database with the OMOP schema, so you should have access to one. Your database access credentials should be kept in .env. An example of the format can be found in /Lettuce/.env.example

Published Images

Development Docker images for the Lettuce project are available on GitHub Container Registry (GHCR):

• Registry: ghcr.io/health-informatics-uon/lettuce
• Weights Image (pre-loaded LLaMA-3.1-8B weights):
  • dev-weights-llama-3.1-8B-sha-<hash> (e.g., -sha-a1b2c3d)
  • dev-weights-llama-3.1-8B-edge (latest)
  • Pull: docker pull ghcr.io/health-informatics-uon/lettuce:dev-weights-llama-3.1-8B-edge
• Base Image (lightweight, no weights):
  • dev-base-sha-<hash> (e.g., -sha-a1b2c3d)
  • dev-base-edge (latest)
  • Pull: docker pull ghcr.io/health-informatics-uon/lettuce:dev-base-edge

See GitHub Packages for all tags.

Contact

If there are any bugs, please raise an issue or email us.