Placer (+ Iliad, the demo EHR it runs on)

A hackathon project for agentic discharge-disposition planning. The idea: once a care team decides where a hospitalized patient should go after discharge (home, skilled nursing, assisted living, inpatient rehab, hospice, …), a lot of slow human legwork stands between that decision and the patient actually leaving — calling SNFs for open beds, calling family for preferences, ordering SNF-required labs (COVID tests), getting consults (PM&R) to qualify for rehab. Patients wait extra days in the hospital for this.

Placer predicts the likely disposition from the patient's chart early and works those barriers in the background (computer use + phone calls), so the paperwork and coordination are already moving by the time the team commits.

Iliad is the general-purpose demo EHR that Placer runs on top of. Placer is built in parallel as its own product; inside the Iliad UI it surfaces as a per-patient Placer tab (predictions, care tasks, call log, and a chat thread with the care team).

This repo

• backend/ — Iliad's backend: a lightweight, Epic-like EHR with a clean REST API over a single SQLite database. It mimics the core EHR functions (patients, encounters, notes, labs, meds, orders, problems) and hosts the Placer-facing constructs (predictions, care-task worklist, facilities, call logs, provider↔Placer chat). Everything is synthetic.
• frontend/ — Iliad's UI: a lightweight, Linear-styled app over that API — patient worklist + chart, orders/notes/labs with their write actions, facility search, and the per-patient Placer tab.

See backend/README.md for the full API reference, frontend/README.md for the UI, and CLAUDE.md for architecture and conventions.

Quickstart

cd backend
python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt

python -m iliad.cli reset          # build + seed the SQLite database
python -m iliad.cli serve --reload # API at http://localhost:8000  (docs at /docs)

In a second terminal, start the UI:

cd frontend
npm install
npm run dev                        # UI at http://localhost:5173

Then explore the API directly if you like:

curl 'http://localhost:8000/patients?admitted=true'        # the inpatient worklist
curl 'http://localhost:8000/patients/hero-a-stroke/chart'  # one-call chart snapshot

Reset to the seed state anytime (great between demo runs):

curl -X POST http://localhost:8000/admin/reset      # or: python -m iliad.cli reset

Demo patients

Four active inpatients are seeded with deep charts (prior inpatient and outpatient encounters, long realistic H&Ps, daily progress notes, discharge summaries, family-communication notes) primed for disposition prediction:

ID	MRN	Sketch	Likely disposition
hero-a-stroke	MRN90001	78F, stroke, lives alone in a 2nd-floor walk-up, daughter out of state	SNF
hero-b-chf	MRN90002	66M, CHF exacerbation improving, lives with spouse in single-story home	Home + home health
hero-c-hospice	MRN90003	84F, metastatic cancer, comfort-focused (DNR), daughter is caregiver	Hospice
hero-d-ambiguous	MRN90004	71M, pneumonia + COPD, social situation undocumented	Undetermined (tests reasoning)

Status

• Iliad backend: demo EHR API + seed/reset + Placer chat thread
• Iliad frontend: demo UI with the per-patient Placer tab
• Placer agents: disposition prediction + background task execution

Data & safety

All clinical data is synthetic (hand-authored demo patients; the synthetic-examples/ FHIR set provided by Abridge is kept for reference but no longer imported). No real PHI is present, and none should ever be committed.