Inspiration

US hospitals are under a chronic nursing workforce shortage: the 2025 NSI national staffing report reports 16.4% Registured Nurse (RN) turnover and a 9.6% vacancy rate. [1] At the same time, 39.9% of RNs intend to leave the workforce or retire within the next five years, with stress/burnout among the top reasons. [2]

Nurses spend more than 1/3 of their shifts documenting instead of caring for patients. [3] [4] Yet, 94% of shift handoffs have errors, and more than 1 in 4 nurses say important information is always missed. [5]

We built ShiftHero to solve this problem: reduce documentation load but also improve accuracy during the most failure-prone transition in a nurse’s workflow: handoff form-filling.

ShiftHero

ShiftHero is an AI-powered assistant that automates nursing handoffs and documentation.

It listens to nurse(-to-patient) conversations, fills out handoff forms and creates structured summaries in real time with minimal hallucinations. It also surfaces patient risk insights with the CDC's Social Vulnerability (CDC-SVI) Dataset. [6]

How We Built It

ShiftHero is a full-stack web app built with Python, Next.js, and React.

Capture + Speech-to-Text

When a patient encounter begins, the app records the conversation. We run speech-to-text (Whisper) to produce a timestamped transcript with speaker separation (nurse vs patient), which becomes the canonical input to the pipeline.

Form-Filling LLM Loop

We pass the transcript into an LLM (GPT-4o), which fills out the handoff form. Instead of generating one long note, the model iterates field-by-field, returning strict JSON that is validated against a predefined schema (Pydantic/JSON Schema), significantly reducing hallucinations.

Retrieval-Augmented Verification (RAG)

After initial extraction, a RAG verification pass re-queries the transcript (and any retrieved reference context) to explicitly justify key entries and flag uncertain or missing fields. Those low-confidence cells are highlighted in the UI so the nurse can confirm in seconds. This further reduces hallucinations.

Dashboard + Workflow UX

The frontend (Next.js/React) shows the most recent handoff immediately after completion, alongside a searchable history. The handoff form renders as a grid with missing/uncertain cells highlighted, plus a side panel for “what to ask next.”

Social Vulnerability Awareness

We store CDC/ATSDR Social Vulnerability Index (SVI) locally and map patient ZIP/location to county/tract-level signals. Those geospatial risk factors are retrieved and injected into the model context so the handoff includes practical “things to watch out for” (transportation, housing instability, socioeconomic stressors) that are hard for nurses to infer mid-shift.

Using both the transcript gaps (from schema validation + RAG) and SVI signals, ShiftHero generates targeted follow-ups directly in the dashboard (e.g., “Do you have reliable transportation to pick up these prescriptions?”) so nurses close high-risk omissions before shift change.

Citations

[1] NSI Nursing Solutions — 2025 NSI National Health Care Retention & RN Staffing Report (PDF). https://www.nsinursingsolutions.com/Documents/Library/NSI_National_Health_Care_Retention_Report.pdf

[2] National Council of State Boards of Nursing (NCSBN) — Workforce study highlights (intent to leave/retire; burnout/stress). https://www.ncsbn.org/news/ncsbn-research-highlights-small-steps-toward-nursing-workforce-recovery-burnout-and-staffing-challenges-persist

[3] Hendrich et al. — A 36-hospital time and motion study: how do medical-surgical nurses spend their time? (documentation 35.3%). https://pubmed.ncbi.nlm.nih.gov/21331207/

[4] U.S. Surgeon General — Addressing Health Worker Burnout (Figure: nurses spend up to 41% of time on EHR/documentation). https://www.ncbi.nlm.nih.gov/books/n/surghealthwwa/pdf/

[5] BMC Nursing (2025) — Quantifying acute care nurses’ experiences of patient handoffs during shift change (6.3% “always” accurate/complete/pertinent; >1 in 4 report info often missed). https://link.springer.com/article/10.1186/s12912-025-03802-6

[6] CDC/ATSDR (GRASP) — Social Vulnerability Index overview / dataset. https://www.atsdr.cdc.gov/place-health/php/svi/index.html