- CLEAR [Clinician-Led, Easy-to-read, Accessible Reaffirmed Instructions]
- Question, problem, or opportunity: [Better patient's and clinician's experience during ER discharge]
- Data, inputs, or evidence: [Synthetic representative data]
- Expected output: [Demonestrate a working workflow]
- Tools and stack: [Python, Foundary llms, streamlit, GitHub]
- Team lead: [Ramzi Alsallaq, ralsallaq]
- Team members and roles: project-management/team.md
- Communication: Slack
Naming the tools and stack early helps the team lead create useful roles and divide work realistically. It is fine to revise this section as the project develops.
After a clinic visit, often late in the day, at the end of a long and stressful encounter, families are handed a large volume of instructions verbally and on paper in one compressed window. In pediatric hematology-oncology and sickle cell care, that volume is high and consequential: opioid and supportive-medication schedules, hydration and home-care measures, and specific "when to call" or "when to return to the ER" thresholds. This is precisely the moment families are least able to absorb and retain detail, and they leave without a clear, durable, organized version of the plan they can return to at home when they are calmer and actually trying to follow it. For the large share of families whose primary language is Spanish, a language and health-literacy barrier compounds the problem.
A printed after-visit summary technically "gives the instructions," but it is long, dense, English-first, and written above a usable reading level, so it fails at the real job: making the plan followable at home.
Prototype an interactive, bilingual home-care companion that presents a family's specific discharge plan in plain language (target ~5th–6th grade), organized by their actual issue (e.g., sickle cell pain managed and discharged; fever/neutropenia precautions). Crucially, the medical content is not generated by the model. The clinician selects pre-built, specialty-vetted home-care modules at discharge, and the visit's actual orders (medications, doses) are populated into them; the app's role is to match, simplify, translate, and present, never to originate clinical instructions.
A working prototype demonstrating the full clinician-in-the-loop pipeline on three vetted condition modules authored by the clinical proposer: Content library: three specialty-vetted home-care modules: (a) sickle cell vaso-occlusive pain after discharge, (b) fever/neutropenia precautions and when to return, and (c) post-chemotherapy nausea/vomiting and home hydration, authored by a board-certified pediatric hematology-oncology attending. This pre-approved content is the safety foundation; the app selects and presents it, never generates it. Personalization layer: the clinician selects the relevant module(s) at discharge; the visit's actual medications and doses are populated into the plan. Critical items (doses, return thresholds, contact triggers) are rendered as structured, verbatim-preserved fields. Presentation layer: an interactive, mobile-friendly plan: plain-language sections the family can expand, a checkable medication schedule, supportive-care steps, and a prominent "call now / return now if…" panel. Delivered in parallel in English and Spanish.
| When | Focus | Expected outcome |
|---|---|---|
| Day 1 | Agree on the question, inputs, stack, roles, and first tasks | A shared plan and a first small change in the repository |
| Day 2 | Build, test, and compare approaches | A working result or clear evidence about what does not work |
| Day 3 | Stabilize, document, and present | A demo or handoff with methods, limitations, and next steps |
The goal is not a perfect production system. The goal is a clear, honest, useful result that the team can explain and others can build on.
