SWIMS-CONNECT

Inspiration

Child-protection concerns are often first noticed by relatives, teachers, neighbours, community leaders, and frontline workers. Reporting can be delayed by unfamiliar channels, incomplete information, repeated paperwork, and fragmented follow-up.

Ghana already uses SWIMS—the Social Welfare Information Management System—built on the open-source Primero platform. We created SWIMS-Connect to make this system easier to reach through WhatsApp while keeping Primero as the authoritative case-management platform.

SWIMS-Connect won the UNICEF StartUp Lab Reducing Child Labour and Trafficking through Systems Innovation Challenge in June 2026.

What it does

SWIMS-Connect supports community reporting and authorised social-worker casework through WhatsApp.

Community members can:

report a child-protection concern using text, voice, or an image;
report without creating a SWIMS account;
decline follow-up contact and remain anonymous; and
receive a SWIMS Case ID after submission.

Authorised social workers can securely sign in and:

retrieve cases permitted by their Primero role;
review case details and the original report;
generate and schedule caseload reports;
draft assessments and case plans for review;
find real social-service providers;
record referrals, service delivery, and follow-ups; and
request or complete closure according to their permissions.

The agent presents casework drafts for review and does not save them until the worker explicitly approves.

How we built it

The conversational assistant is a Python LangGraph coded agent deployed on UiPath Automation Cloud. Google Gemini provides natural-language and media understanding, while typed tools perform controlled operations against the Primero REST API.

A Node.js WhatsApp gateway connects users to the deployed agent through the UiPath TypeScript conversational SDK. It handles message routing, consent, media transport, secure worker-login links, and authenticated worker context.

UiPath Orchestrator manages the agent package, release, secrets, configuration assets, jobs, and traces.

A deployed UiPath Maestro Case provides persistent overdue-work monitoring. For eligible worker-filed cases, Maestro tracks assessment, case-plan, referral, follow-up, and closure-review deadlines. Before sending a reminder, the system verifies the case’s current state and exact due dates in Primero.

Worker credentials and Primero sessions are never exposed to the language model. A short-lived, sender-bound token carries the worker context and is removed before model processing.

The solution was developed using Claude Code and UiPath for Coding Agents.

Challenges we ran into

Connecting WhatsApp to a conversational UiPath agent while preserving conversation state.
Carrying authenticated worker context without exposing credentials to the model.
Mapping natural-language reports into valid Primero fields and lookup values.
Preventing fabricated Case IDs by accepting identifiers only from Primero.
Supporting anonymous intake alongside role-protected worker operations.
Handling voice notes, images, retries, duplicate messages, and gateway restarts.
Coordinating long-running Maestro monitoring with live Primero workflow information.
Keeping scheduled reports and overdue reminders within the signed-in worker’s permissions.

Accomplishments that we're proud of

Deployed a working conversational coded agent on UiPath Automation Cloud.
Created real Primero cases from natural-language, voice, and image reports.
Implemented anonymous community reporting with a mandatory consent gate.
Built secure, role-scoped social-worker access.
Added assessment, case-plan, referral, service-delivery, follow-up, and closure tools.
Implemented 13 on-demand and scheduled report types.
Deployed a UiPath Maestro Case for persistent deadline monitoring.
Connected confirmed Primero closure to automatic Maestro cancellation.
Ensured workers review drafts before anything is saved.
Won the UNICEF StartUp Lab child-labour and trafficking systems challenge.

What we learned

A trustworthy casework agent requires much more than a capable language model. Identity, permissions, consent, deterministic writes, auditability, safeguarding, and human review are equally important.

We also learned that Maestro and Primero have complementary responsibilities. Maestro provides long-running visibility and deadline monitoring, while Primero remains authoritative for protection records, workflow state, and role-based access.

Conversational interfaces can reduce administrative effort, but they must strengthen professional judgement rather than replace it.

What's next for SWIMS-CONNECT

We are preparing for a pilot in Ghana focused on safeguarding review, frontline-worker feedback, operational reliability, and measuring reductions in administrative workload and missed follow-ups.

Planned improvements include deeper Maestro lifecycle coordination, additional human-review workflows, local-language support, stronger offline reporting, expanded service-directory integration, and deployment monitoring.

Because Primero is configurable and used in more than 80 countries and territories, SWIMS-Connect can be adapted to other deployments through country-specific forms, languages, roles, protection concerns, referral directories, and governance requirements.