Quickstart

The example notebook provides a runnable walkthrough using sample input data:

Open examples/run_examples.ipynb on GitHub

Most external workflows should start with these import paths:

from matchminer_ai import load_config
from matchminer_ai.trials import summarize_trials
from matchminer_ai.patients import summarize_patients
from matchminer_ai.embedding import embed_for_matching
from matchminer_ai.matching import (
    exclusion_criteria_check,
    exclusion_criteria_check_with_llm,
    generate_candidate_matches,
    score_match_quality,
    score_match_quality_with_llm,
)

Before starting, review the data requirements for the trial and patient input tables expected by the workflow.

The patient-centric workflow has these main steps:

1. Turn trial descriptions and eligibility criteria into structured trial summaries with summarize_trials.
2. Turn patient notes into cancer history summaries with summarize_patients.
3. Convert the trial and patient summaries into embeddings with embed_for_matching so they can be compared.
4. Find likely patient-trial pairs with generate_candidate_matches.
5. Score how reasonable each patient-trial match looks with either score_match_quality or score_match_quality_with_llm.
6. Screen matched patient-trial pairs for possible exclusion criteria with either exclusion_criteria_check or exclusion_criteria_check_with_llm.