Zero-Configuration Retention Intelligence · HIV Programme Intelligence

Your programmes are data rich.
Insight poor.

8,000-line production-grade analytics. Drop standard DHIS2 and EMR exports directly into the platform - no data cleaning, no infrastructure changes, zero patient data liability.

→ Launch live synthetic demo View security architecture

No data sharing agreement needed · Synthetic data available instantly · No sign-up

23,144
Adults in development cohort
0.806
Model AUC · temporal holdout
55+
PEPFAR countries · target market
<5 min
Time to first insight
Credibility signals
Open code · github.com/Kchinthala15
In discussion · PEPFAR implementing partners & ministries of health
Explainable AI · SHAP-powered predictions
Temporal validation · AUC 0.806 (95% CI 0.774–0.837)
The platform

A real tool. Not a concept.

Live and publicly accessible. Upload your programme data - in minutes, SmartDaaS generates a ranked patient list and a weekly outreach plan fitted to your actual staff capacity.

SmartDaaS  ·  Outreach Optimiser
Model loaded  ·  AUC 0.806
Start here
Home
Patient Risk
Programme
Outreach Optimiser
Executive Report
Validation
Facility Intelligence
Cohort Intelligence
Model Transparency
This week's outreach plan
2 workers · 5 days · 3h/day · 30h total capacity
87.8% high-risk coverage
847
Total patients
124
High risk
45
To contact
13.8
Est. prevented
Patient priority list - highest outreach ROI first
RankPatient IDRiskScorePrimary focus
1PT-0041HIGH83.4%Side effects → regimen review
2PT-0039HIGH83.6%Side effects → regimen review
3PT-0031HIGH96.8%Prior interruption → re-engagement
4PT-0018HIGH74.7%TB positive → TB/HIV integration
5PT-0009MED61.2%CD4 below threshold → urgent review
The challenge
"Out of hundreds of at-risk patients and limited outreach staff - who do we contact first?"

Every week, HIV programme teams across sub-Saharan Africa face this question with no tool to answer it. The data exists. The operational guidance does not.

01
1 in 3 patients is lost to follow-up across sub-Saharan Africa - not from lack of data, but from lack of clarity on where to act.
02
$1.4B in PEPFAR funding is at risk annually from preventable patient loss. M&E teams spend more time on reports than on action.
03
No existing tool converts patient risk scores into a capacity-constrained weekly action plan fitted to actual staff hours. SmartDaaS does.
System compatibility

Works with the data your team already pulls.

SmartDaaS natively recognises column names from the major national HIV data systems across sub-Saharan Africa. No reformatting. No data engineering.

Kenya NASCOP
Kenya National AIDS & STI Control Programme
Columns from NASCOP's patient-level export are mapped automatically. Upload your monthly CSV - no renaming required.
✓ Native column mapping
Uganda DHIS2
District Health Information System 2
Connects to DHIS2 tracker programmes and reads tracked-entity records directly - patient attributes, enrollments and events - so no CSV export is needed. Aggregate DHIS2 data is not used; the model scores at the patient level.
✓ Native column mapping
Malawi HMIS
Health Management Information System
Malawi HMIS patient-level fields are pre-mapped. Upload the same export your M&E Advisor already runs monthly.
✓ Native column mapping
PHIA Surveys
Population HIV Impact Assessment
PHIA 2020+ wave variables (MPHIA, THIS, UPHIA) are recognised for population-level analysis - a plausibility check, not patient-level validation.
✓ Native column mapping
Generic EMR export
Any patient-level CSV
109 column aliases cover common naming conventions across EMR systems used by PEPFAR implementing partners. If your column isn't recognised, it's flagged - not silently dropped.
✓ 109 column aliases
No API required
Zero integration work
No IT sign-off. No data pipeline. No new systems. SmartDaaS is a CSV upload - just like the monthly report your team already sends.
✓ Instant evaluation
DHIS2 Native API
Any DHIS2 instance worldwide
Connect directly to your DHIS2 instance via the DHIS2 Web API - no CSV download required. SmartDaaS pulls patient data automatically, maps columns natively, and runs the full analysis pipeline.
✓ Native API connection
Evaluating SmartDaaS requires no data sharing agreement. The live platform runs on synthetic data - upload the sample dataset to see a full outreach plan generated in real time, before sharing a single row of patient data.
How it works

From data upload to actionable plan in minutes.

01 - Upload
Your existing export
Upload the CSV your team already pulls every month. No new data collection. No new systems. SmartDaaS works with what you have - and recognises your column names natively.
02 - Analyse
AI scores every patient
The risk engine ranks every patient by adherence risk - with transparent, SHAP-explainable reasons your clinical team can interrogate.
03 - Act
Your weekly action plan
Enter your outreach capacity - staff count, days, hours. SmartDaaS builds the plan: exactly who to contact, in what order, this week, fitted to your actual resources.
Security & Architecture

Built for institutional data governance.

Three architectural pillars built around how ministries and implementing partners actually handle patient data.

Pre-Mapped Schema Ingestion
Stop writing data cleaning scripts. Native column aliasing standardises fragmented country-level exports automatically.
NATIONAL_SURVEYS
Direct zero-prep compatibility: MPHIA, THIS, UPHIA population-based surveys - no reformatting.
HMIS_PIPELINES
Pre-configured variable mapping for Kenya NASCOP, Uganda DHIS2, and Malawi HMIS CSV exports. Now supports direct DHIS2 Web API connection - no CSV export required.
COLUMN_ALIASES
109 aliases covering EMR naming conventions across PEPFAR implementing partners. Unrecognised columns are flagged - never silently dropped.
Zero Data Liability
Three deployment models to fit your governance. Federated and self-hosted keep patient data inside your infrastructure entirely. Ethics and data-transfer review remain your institution's determination.
STATELESS_PROCESSING
Retains zero Protected Health Information or patient PII. All computation is ephemeral - nothing persists post-session.
SHA256_HASHING
Uploaded file names are SHA-256 hashed before logging, so no filename metadata enters the audit trail. The platform expects de-identified input and does not itself de-identify.
AUDIT_LOGGING
Aggregate-only audit logging (event, timestamp, cohort size, mean risk - no patient-level field), with a per-session login-attempt limit and automated 4-hour session timeouts.
Real-World Data Resilience
Frontline health data is messy. SmartDaaS doesn't break - it adapts, grades, and recalibrates.
DQ_GRADE
Automated data quality tiering profiles incoming cohorts and assigns a reliability grade before scoring begins.
CAL_METHOD
Recalibration adjusts the modelling pipeline to the data tier, and the reliability grade travels with the output so you know what the data supports.
SHAP_15_FEATURES
No black boxes. Every patient-level risk score includes a full SHapley Additive exPlanations breakdown - exact clinical factors, fully auditable.
smartdaas · security_spec.py
# SmartDaaS security architecture · federated and self-hosted deployment

PHI_RETAINED         = False       # stateless processing layer
PII_STORED           = False       # zero patient data liability
FILENAME_HASHING     = "SHA-256"    # filename only · expects de-identified input
SESSION_TIMEOUT_HR   = 4           # automated inactivity lockout
LOGIN_ATTEMPT_LIMIT  = 5           # per session · not a substitute for network controls
AUDIT_LOG            = "supabase"   # enterprise-grade · full traceability
ETHICS_REVIEW        = "your_determination"  # we make no claim it is unnecessary
CROSS_BORDER_TRANSFER = False       # federated / self-hosted only · hosted instance is US-based

>>> deployment_options()
✓ Federated · validation runs behind your firewall, returns aggregate metrics only
✓ Self-hosted · Docker container inside your infrastructure
✓ Hosted · US instance, for evaluation on synthetic or de-identified data
DEPLOY: FEDERATED · SELF-HOSTED · HOSTED
OPEN CODE github.com/Kchinthala15/smartdaas-hiv-validation

Every number here is reproducible.

The model, the code that trained it, and the results are all public. One script reproduces every performance number on this page. Rather than take the figures on trust, implementing partners and ministries can run them.

View the code and model card →
0.806
AUC on temporal holdout Trained on patients who started ART up to Sept 2016, then tested on 6,942 who started after - a stricter test than in-sample accuracy. 95% CI 0.774–0.837.
23K
Patients in training dataset Developed on 23,144 adults from a 27,288-patient programme deposit, then tested on 6,942 patients who initiated ART after the training window - not a small academic sample.
15
Clinical features · SHAP-explained Every prediction is backed by an explainable feature contribution - so clinicians see exactly why a patient was flagged high-risk.

Every number on this page is reproducible from one script.

The model, the training code, the validation results and the limitations are public.

github.com/Kchinthala15/smartdaas-hiv-validation

Ready to turn your data
into decisions?

Built for HIV programmes, implementing partners, and public health systems.

In discussion with PEPFAR implementing partners and ministries of health across sub-Saharan Africa.
No disruption. No new data collection. Evaluate on synthetic data before sharing a single row.

Request a pilot Try the live platform →
No data sharing agreement No IT approval needed Evaluate in under 5 minutes
Platform live · synthetic demo, no sign-up