the problem

Lucero Health runs a "whole person care" model in Harris County: community health workers try to follow up with patients after a clinic visit to make sure they actually get the food, housing, transportation, or other community resources they were referred to, not just the medical treatment itself. The problem is scale: a small team of health workers can't manually check in with every patient, so referrals quietly go unfollowed and the visit's impact stops at the clinic door.

what we built

A two-sided GenAI platform. On the patient side, an agentic follow-up system reaches out after a visit over Twilio, so patients don't need to install anything. On the resource side, RAG retrieval and ranking match each patient's referral to the right community resource from Lucero's database. I owned defining the follow-up completion rate (the metric we used to tell whether the pilot was actually closing the loop) and worked with the team on responsible AI deployment: defining the categories of error states the agent could hit, tuning the pipeline against each one, and keeping a human review step before anything reached a patient.

results

We piloted across 2 Harris County clinics, reaching 78 patients, and cut missing post-visit follow-ups by 20% over the course of the pilot.