IncidentIQ

IncidentIQ is designed for those who report incidents, not just those who analyze thee. Enabling fast, low-friction reporting while delivering insights to help prevent future harm.

Comment

IncidentIQ

AI-powered safety reporting for nursing homes

Inspiration

Up to 96% of workplace near-misses go unreported in healthcare settings. But the burden doesn't just fall on staff who are afraid to report — it falls on the administrator who receives each report and faces hours of manual work: reading through free-text descriptions, deciding OSHA severity, hand-typing four narrative sections into Form 301, cross-referencing injury logs, and tracking workers' comp timelines, all while managing a full facility.

We asked: what if AI could absorb all of that cognitive load the moment a report comes in?

What it does

IncidentIQ is an AI-first incident management platform for nursing home administrators and frontline staff. The moment a report is submitted, Claude gets to work — so by the time an administrator opens a case, most of the hard thinking is already done.

For administrators — this is where AI changes everything

  • Instant triage on every case. The moment a report is submitted, Claude analyzes the free-text description and structured data to assign a severity level (high / medium / low), flag OSHA recordability, write a plain-English summary, and generate a prioritized list of recommended next actions. Administrators no longer have to read every report from scratch and make judgment calls alone — Claude surfaces the signal.

  • OSHA 301 drafts itself. Form 301 requires four specific narrative sections that typically take 20–30 minutes to write per incident. Claude pre-fills all four sections — "What happened," "What was the object/substance," "What bodily part was affected," and "What was the task" — using the submitted report. The administrator reviews, edits if needed, and prints. A form that used to take half an hour takes two minutes.

  • OSHA Log stays current automatically. Every recordable case flows into the OSHA 300 Log in real time with correct classification. Administrators can export the full year as CSV or PDF at any time — no manual data entry, no end-of-year scramble.

  • Case management with full audit trail. Assign reviewers, send notes to employees, track workers' comp status, and close cases — every action is timestamped and logged. If OSHA ever asks, the paper trail is already there.

  • Employee notifications built in. When a reviewer is assigned or a case status changes, employees are notified automatically. Administrators can also send manual case updates with one click.

For frontline staff

  • Guided 2-minute report form with contextual tips and near-miss support
  • Personal portal to track case status and see notes from administrators

How we built it

Layer Technology
Frontend React 18 + Vite + Tailwind CSS v4
AI triage & OSHA 301 drafting Claude (Anthropic) via n8n automation
State management React Context + localStorage
Design Custom component system, healthcare-focused

Claude is called via n8n automation workflows — triggered on new case submission, returning structured JSON that populates the case record immediately. The full demo runs without a backend.

The AI workflow in detail

When a report is submitted:

  1. n8n catches the submission event and sends the case data to Claude via the Anthropic API
  2. Claude returns a structured triage object: severity, OSHA recordability flag, a 1–2 sentence summary, and 3–5 recommended actions
  3. That triage data is attached to the case and rendered immediately when the administrator opens it
  4. When the administrator opens the OSHA 301 form, Claude is called again with the full incident details to draft all four required narrative fields
  5. The administrator sees a pre-filled form they can edit and print — not a blank page

The administrator goes from "new report in queue" to "ready to act" without reading a wall of text or making cold judgment calls.

Challenges we ran into

Getting Claude's triage output to feel trustworthy — not like a black box — required careful prompt design. We structured the output explicitly: severity must be one of three values, OSHA recordability is boolean with reasoning, and recommended actions are written as specific next steps, not vague suggestions. Administrators can trust the output because it's transparent about what it found and why.

Accomplishments that we're proud of

The OSHA 301 auto-draft is the feature we're most proud of. Every nursing home doing this correctly is spending dozens of administrator hours per year writing these forms. We built a workflow where Claude writes a compliant, facility-specific draft in seconds. That's real time returned to people doing hard work in understaffed environments.

We're also proud of the audit trail design — every AI action, every status change, every note is logged with timestamp and actor. The system is defensible the moment something is submitted.

What we learned

AI in compliance workflows has to be a co-pilot, not autopilot. Every Claude output in IncidentIQ is editable and reviewable before anything gets signed or submitted. The administrator stays in control — AI just eliminates the blank-page problem and surfaces what matters.

What's next for IncidentIQ

  • Pattern analysis: Claude identifying recurring hazards by location, shift, or job role — turning incident data into preventive action
  • Real authentication and multi-facility support
  • SMS/email delivery for staff notifications
  • EHR integration for employee record pre-fill
  • Mobile-optimized near-miss capture for reporting on the floor

Built With

Share this project:

Updates