InterSystems Programming Contest: AI Agents for FHIR

1. Smart Patient Summary Generator
   1. What it does: Creates a concise, clinician-friendly summary of a patient from FHIR data, including conditions, medications, allergies, recent encounters, labs, and care plans.
   2. FHIR resources: Patient, Condition, MedicationRequest, AllergyIntolerance, Observation, Encounter, CarePlan.
   3. Platform features to use: FHIR API, FHIR SQL Builder, AI Hub. 
   4. 1–2 week MVP: Pick one patient, pull recent FHIR resources, generate: 
      • “current issues”;
      • “recent changes”;
      • “risks / follow-up items”.
   5. Nice twist: Generate different summaries for different roles: ED doctor, care manager, patient, or family caregiver.
2. FHIR Prior Authorization Copilot
   1. What it does: Helps staff prepare prior auth requests by pulling diagnosis, medication history, procedures, and supporting evidence from FHIR data.
   2. FHIR resources: Patient, Coverage, Condition, MedicationRequest, Procedure, Observation, DocumentReference. 
   3. Platform features to use: FHIR SQL Builder, Vector Search, AI Hub, AI Agents. 
   4. 1–2 week MVP: Given a requested medication or procedure, the app: 
      • identifies likely supporting diagnoses;
      • finds relevant recent observations or notes;
      • drafts a justification summary.
   5. Nice twist: Add a “missing evidence” checklist. 
3. Gaps-in-Care Finder for Preventive Screening
   1. What it does: Identifies patients who may be overdue for screenings, vaccines, or chronic disease follow-ups.
   2. FHIR resources: Patient, Immunization, Observation, Condition, Procedure, Encounter.
   3. Platform features to use: FHIR SQL Builder, dashboards/analytics, optional AI Hub. 
   4. 1–2 week MVP: Implement 2–3 rules such as:
      • HbA1c follow-up for diabetes;
      • mammogram screening window;
      • flu/COVID vaccination reminder;
      • blood pressure follow-up.
   5. Nice twist: Use AI to generate outreach messaging tailored to patient preference or risk level.
4. Medication Safety and Interaction Assistant
   1. What it does: Builds a medication review app that flags possible duplication, interactions, allergy conflicts, or adherence issues using FHIR medication and allergy data.
   2. FHIR resources: MedicationRequest, MedicationStatement, AllergyIntolerance, Condition, Observation. 
   3. Platform features to use: FHIR API, AI Hub, Vector Search. 
   4. 1–2 week MVP: Show:
      • active medication list;
      • duplicate therapy detection; 
      • allergy cross-check;
      • plain-language counseling summary.
   5. Nice twist: Use Vector Search over drug guidance content or patient education documents to return context-aware explanations.
5. AI-Powered Care Plan Navigator
   1. What it does: Turns CarePlan and Goal resources into a guided experience for patients or care coordinators.
   2. FHIR resources: CarePlan, Goal, Task, Appointment, MedicationRequest, Observation. 
   3. Platform features to use: AI Agents, AI Hub, FHIR API. 
   4. 1–2 week MVP: The app answers:
      • “What should this patient do next?”;
      • “What tasks are overdue?”;
      • “Which goals are off track?”.
   5. Nice twist: Add an agent that creates suggested Task resources based on changes in observations or missed appointments.
6. Social Determinants and Community Referral Matcher
   1. What it does: Uses clinical and social data to recommend community resources for transportation, food insecurity, housing, or behavioral health support.
   2. FHIR resources: Condition, Observation, QuestionnaireResponse, ServiceRequest, CarePlan, Patient.
   3. Platform features to use: Vector Search, AI Hub, AI Agents.
   4. 1–2 week MVP: Given a patient profile plus a small directory of local services:
      • identify possible needs; 
      • recommend relevant resources; 
      • generate a referral summary.
   5. Nice twist: Store community resources as semantically searchable content and use Vector Search to match needs to services.
7. FHIR Clinical Trial Matcher
   1. What it does: Matches patients to trial eligibility criteria using their coded clinical data and relevant unstructured documents.
   2. FHIR resources: Condition, Observation, MedicationStatement, Procedure, Patient, DocumentReference.
   3. Platform features to use: Vector Search, AI Hub, AI Agents, FHIR SQL Builder. 
   4. 1–2 week MVP: For a few mock trials: 
      • extract criteria; 
      • compare against patient profile;
      • show “likely eligible / maybe / unlikely” with rationale. 
   5. Nice twist: Agent asks follow-up questions like “latest eGFR?” or “prior therapy present?” if criteria are incomplete.
8. Natural Language to FHIR Query Explorer 
   1. What it does: Lets a user ask questions like “Show diabetic patients with A1c above 9 in the last 6 months” and translates that into structured FHIR queries or FHIR SQL.
   2. FHIR resources: Any relevant set, especially Patient, Condition, Observation, Encounter. 
   3. Platform features to use: FHIR SQL Builder, AI Hub. 
   4. 1–2 week MVP: Support 5–10 question patterns:
      • cohort discovery; 
      • recent lab filters; 
      • medication-based populations; 
      • encounter recency. 
   5. Nice twist: Show the generated SQL or FHIR search so judges can see the transparency.
9. Hospital Readmission Risk Workbench
   1. What it does: Builds a prototype that estimates short-term readmission risk from recent utilization, diagnoses, meds, and labs, then proposes interventions.
   2. FHIR resources: Encounter, Condition, Observation, MedicationRequest, CarePlan, Patient.  
   3. Platform features to use: FHIR SQL Builder, AI Hub, dashboards. 
   4. 1–2 week MVP: Use rule-based or lightweight scoring instead of a true ML model:
      • recent admissions;
      • high-risk conditions;
      • missed follow-ups;
      • abnormal labs;
      • polypharmacy.
   5. Nice twist: Generate recommended next steps as Task or CarePlan suggestions. 
10. Conversational FHIR Triage Assistant
    1. What it does: An AI agent asks patients structured triage questions, stores answers as QuestionnaireResponse, and creates a clinician handoff summary.
    2. FHIR resources: Questionnaire, QuestionnaireResponse, Condition, Observation, ServiceRequest, Encounter. 
    3. Platform features to use: AI Agents, AI Hub, FHIR API. 
    4. 1–2 week MVP: The prototype can:
       • ask symptom questions;
       • recommend urgency level;
       • generate a structured handoff note; 
       • create a follow-up task or service request.
    5. Nice twist: Map the conversation into coded FHIR observations where possible.
11. Imaging and Results Follow-Up Tracker
    1. What it does: Tracks abnormal results and ensures they have documented follow-up, making it easier to spot “closed-loop” care failures.
    2. FHIR resources: Observation, DiagnosticReport, ServiceRequest, Task, Encounter, Patient. 
    3. Platform features to use: FHIR SQL Builder, AI Hub, AI Agents. 
    4. 1–2 week MVP: Find abnormal results and show whether there is:
       • a follow-up appointment; 
       • a repeat order; 
       • a task; 
       • a plan in notes.
    5. Nice twist: Use AI to draft an outreach or clinician reminder.
12. Patient-Friendly Lab Explainer 
    1. What it does: Converts lab results and trends into plain-language explanations for patients, with educational context and suggested questions for their doctor.
    2. FHIR resources: Observation, DiagnosticReport, Patient. 
    3. Platform features to use: AI Hub, Vector Search. 
    4. 1–2 week MVP: Select a few lab types such as CBC, A1c, lipids, or CMP and generate:
       • what the result means; 
       • whether it changed over time; 
       • what to ask next. 
    5. Nice twist: Use Vector Search over trusted educational content to ground the explanation. 

*Implementation of these ideas guarantees additional bonuses. 
*Full Technology Bonuses breakdown will be published soon.