Care1OS

• 

Inspiration

Care-home managers work inside a constant stream of staffing, payroll, rota, and compliance pressure. A small rota issue can quickly become a payroll discrepancy, a Working Time Regulation risk, or a missed follow-up for the manager.

Care1OS was inspired by that real operational burden: the need for one calm, practical system that helps care-home teams see what matters, act quickly, and keep evidence attached to decisions.

For OpenAI Build Week, I focused on one high-value workflow: helping managers turn messy workforce evidence into a clear compliance action brief using Codex and GPT-5.6.

What it does

Care1OS is a Laravel-based care-home operations platform covering staff coordination, care records, payroll, compliance, and manager workflows.

For this submission, I added a new GPT-5.6 Workforce Compliance Brief inside the Payroll & Finance dashboard.

The workflow:

1. Runs deterministic Working Time Regulation checks from existing rota and payroll evidence.
2. Removes staff names and resident/care-recipient information before any AI request.
3. Sends only pseudonymous evidence to GPT-5.6 Sol through the OpenAI Responses API.
4. Returns a structured manager brief with:
   • overall risk level,
   • evidence-backed priorities,
   • manager checks to verify,
   • positive signals,
   • and an advisory disclaimer.
5. Does not automatically change payroll, rota, or staff records.

The goal is not to replace the manager. The goal is to help the manager notice the right issues faster.

How we built it

Care1OS is built with Laravel, Blade, jQuery, Bootstrap 3, MySQL, and the OpenAI Responses API.

The Build Week feature was implemented as a manager/admin-only workflow:

• AIComplianceBriefService gathers and anonymises workforce evidence.
• OpenAIService calls the OpenAI Responses API with gpt-5.6-sol.
• The request uses low reasoning effort, strict JSON output, and store: false.
• A Laravel controller exposes a throttled, CSRF-protected endpoint.
• The Payroll & Finance dashboard renders the brief safely using escaped output and jQuery .text().

Codex was used throughout the process to inspect the existing Laravel codebase, design the feature boundaries, implement the service/controller/UI, add tests, package a sanitized judging repository, and document the pre-existing baseline versus the new Build Week work.

GPT-5.6 is used at runtime to transform anonymised compliance evidence into a structured, manager-readable action brief.

Challenges

The biggest challenge was making the AI feature useful without making it unsafe.

Care-home data is sensitive, so the workflow had to avoid sending resident information, staff names, uploads, production records, or raw personal data to the model. That meant designing an evidence format that is useful enough for GPT-5.6 to reason over, while still keeping the privacy boundary tight.

Another challenge was working inside an older Laravel/Blade system rather than a fresh app. The feature had to follow the existing patterns: home-scoped queries, role-based access, CSRF, rate limiting, Blade escaping, and jQuery rendering.

The final challenge was packaging the project for judging. The original platform had real operational history and files, so I created a private sanitized clone with schema-only setup, synthetic demo data, no production uploads, no database dumps, and no secrets.

Accomplishments that we're proud of

I am proud that the feature is practical, not just a demo. It fits naturally into a real manager workflow and keeps the AI output advisory rather than automatic.

I am also proud of the privacy design. The model receives pseudonymous staff references and aggregate compliance evidence, not resident data or staff names.

Finally, I am proud that Codex was used not only to write code, but to help reason through security, testing, repo hygiene, documentation, and submission readiness.

What we learned

I learned that GPT-5.6 is especially useful when paired with deterministic domain logic. The system does not ask the model to invent compliance facts. Instead, Laravel computes the evidence, and GPT-5.6 turns that evidence into a concise, structured action brief.

I also learned that pre-existing projects can be extended meaningfully when the new AI workflow is clearly scoped, documented, tested, and separated from the original baseline.

What's next for Care1OS

Next, I would expand the brief into a broader manager command center:

• trend summaries across multiple weeks,
• follow-up task creation,
• audit-ready manager notes,
• shift anomaly explanations,
• and safer human-in-the-loop workflows for rota planning.

Longer term, Care1OS could become a practical AI operations layer for care homes, helping managers move from reactive admin work to proactive, evidence-led decision-making.

Built With

• codex
• php