AI-Powered AMR Risk Prediction and Infection Control Platform by Team Curelytics for Impact-AI-Thon 2026.
ResisTrack is a clinical decision support system designed to predict Antimicrobial Resistance (AMR) risk within 6 hours of patient admission. By integrating real-time EHR data with advanced machine learning models, the platform aims to reduce inappropriate antibiotic prescribing by at least 30%. It provides clinicians with actionable insights and ward-level visualizations to improve infection control and patient outcomes.
- Cloud Infrastructure: AWS (SageMaker, HealthLake, RDS, Lambda, CDK)
- Backend & ML: Python 3.11, XGBoost, PyTorch LSTM, ClinicalBERT
- Frontend: React.js
- Interoperability: FHIR R4, CDS Hooks
project-root/
|-- roadmap/
| |-- 00_context.md # Project goal, stack, constraints
| |-- 01_roadmap.md # Full phased roadmap
| |-- phases/
| |-- phase_1.md # Secure Data Foundation
| |-- phase_2.md # AI/ML Inference Engine
| |-- phase_3.md # Clinical Integration & Alerts
| |-- phase_4.md # Dashboard, Reporting & UX
| |-- phase_5.md # Validation, MLOps & Hardening
|-- modules/
| |-- m1_1_aws_infra_baseline/ # VPC, CDK, KMS, IAM
| |-- m1_2_hospital_connectivity/ # VPN, TLS, API Gateway
| |-- m1_3_hl7_fhir_transformer/ # Mirth Connect, HL7->FHIR
| |-- m1_4_fhir_ingestion_tokenization/ # Lambda, PHI tokenization
| |-- m1_5_data_storage_layer/ # HealthLake, RDS, S3
| |-- m1_6_audit_logging/ # CloudTrail, compliance
| |-- m2_1_feature_engineering/ # 47 tabular features
| |-- m2_2_temporal_features/ # LSTM tensors
| |-- m2_3_nlp_clinicalbert/ # ClinicalBERT fine-tuning
| |-- m2_4_xgboost_model/ # XGBoost training
| |-- m2_5_lstm_model/ # PyTorch LSTM
| |-- m2_6_ensemble_calibration/ # Ensemble + SHAP
| |-- m2_7_sagemaker_endpoint/ # Real-time inference
| |-- m3_1_smart_fhir_auth/ # OAuth, SMART launch
| |-- m3_2_cds_hooks_service/ # CDS Cards, EHR integration
| |-- m3_3_override_monitoring/ # Override rate tracking
| |-- m3_4_notification_dispatch/ # SNS notifications
| |-- m4_1_dashboard_shell/ # React app, auth
| |-- m4_2_ward_heatmap/ # Ward risk visualization
| |-- m4_3_patient_timeline/ # Patient risk + SHAP panel
| |-- m4_4_pharmacy_ic_views/ # Role-specific views
| |-- m4_5_stewardship_reports/ # PDF/CSV reports
| |-- m5_1_clinical_validation/ # Model validation gate
| |-- m5_2_mlops_retraining/ # Auto retraining pipeline
| |-- m5_3_cicd_pipeline/ # CodePipeline + CodeBuild
| |-- m5_4_monitoring_alerting/ # CloudWatch, alarms
| |-- m5_5_compliance_docs/ # HIPAA, FDA PCCP, model card
|-- prompts/ # Reusable prompt templates
|-- logs/ # Decision records
|-- ResisTrack_Agent_Rules.md # Agent rules for AI models
|-- ResisTrack_Roadmap.md # Source roadmap document
Start with roadmap/00_context.md for project context, then roadmap/01_roadmap.md for the full plan. Each module in modules/ contains a spec.md with scope, dependencies, and a done-when checklist. Implementation code is located in impl/ and tests in tests/ within each module directory.
| Phase | Title | Duration | Modules |
|---|---|---|---|
| 1 | Secure Data Foundation | 3 Weeks | 6 |
| 2 | AI/ML Inference Engine | 4 Weeks | 7 |
| 3 | Clinical Integration & Alerts | 2 Weeks | 4 |
| 4 | Dashboard, Reporting & UX | 3 Weeks | 5 |
| 5 | Validation, MLOps & Hardening | 3 Weeks | 5 |
- HIPAA Compliance: All data handling must adhere to HIPAA regulations.
- Data Privacy: No PHI is permitted outside the VPC.
- Clinical Safety: The model provides decision support and never prescribes medications directly.
- Validation: Clinical validation is required before any model deployment.
Curelytics, Impact-AI-Thon 2026