AnticiPatient

Problem statement

Hospital selection in an urgent-care moment is a high-stakes, multifaceted decision and almost nobody is equipped to make it well under pressure. In the seconds after something goes wrong, a person typically has to resolve several things at once:

• Symptom triage: is this an ER-now situation, an urgent care, or something that can wait, and which clinical specialty matches the presentation.
• Hospital fit: which nearby hospitals actually have strength in that specialty, and which are rated well for the kind of care needed.
• Coverage and cost: which facilities are in-network, what copays and downstream costs look like, and how affordability influences the safest choice.
• How to get there: whether to drive, be driven, call an Uber, or call EMS, given traffic, helping hands, and the patient's own ability to move.
• What to do in the meantime: real-time guidance on pre-hospital care (applying pressure on an open wound, recovery position, keeping a stroke patient awake) while en route.

Even informed insiders in the medical system are bottlenecked on the last mile: real-time ER wait times, live traffic between the patient's exact location and each hospital, and transportation availability are almost never collated in one place. Everyone, expert or not, ends up piecing this together from phone calls, static hospital websites, and gut instinct, which is exactly the wrong workflow for a minutes-matter decision.

Solution

AnticiPatient fuses two streams of information into one decision surface:

1. Real-time hospital signals: ER wait times, specialty availability and reputation, network/insurance coverage, and live travel conditions to each candidate facility.
2. Patient and incident context: symptoms, location, mobility, access to a driver or helping hands, insurance, and any relevant medical history.

From those inputs, the platform produces a single, ranked recommendation that answers four questions at once:

• Where to go (the hospital best matched to specialty, wait time, coverage, and travel time).
• How to get there (self-drive, ride, or EMS, based on condition severity and available helping hands).
• What to do now (step-by-step pre-hospital care for the specific presentation).
• How to navigate (turn-by-turn directions to the recommended hospital, updated live for traffic).

Passive safety layer

A toggleable background feature allows AnticiPatient to automatically contact emergency services if the user becomes unresponsive during an incident, using phone-side activity and health signals as the trigger. Opt-in and opt-out are a single switch, so users stay in control of when the phone is allowed to place a call on their behalf.

My role and contributions

I worked across the stack to ship data ingestion, APIs, and client views, with the product framing above as the north star: every feature was judged against whether it reduced the number of tabs and phone calls a patient or caregiver had to juggle in an emergency. After accelerator selection, the project pivoted from a startup track to an open-source initiative, which shifted priorities toward transparency, documentation, and community adoption so the decision engine could be audited and extended by clinicians and contributors outside the core team.

Technical approach

The platform combines scheduled and event-driven ingestion of hospital-facing signals with normalized storage and HTTP APIs for the web client. The recommendation engine joins the live data against a per-incident context payload (symptoms, location, coverage, mobility) to produce the ranked hospital + transport + pre-hospital-care output in a single response. The UI focuses on legible comparisons (waits, specialties, logistics) rather than dense dashboards, because the end user is often stressed and one-handed.

Ingestion:        scheduled pulls + event hooks for ER wait times,
                  specialty availability, insurance network, traffic
Normalization:    hospital-agnostic schema, JSON APIs for the client
Recommendation:   rank on (specialty match x wait time x travel time
                  x in-network x incident severity)
Outputs:          hospital pick, transport mode, pre-hospital steps,
                  live directions
Passive safety:   optional unresponsive-user detection -> auto 911

Results

• Selected into Columbia University's CS3 VALIDATE Accelerator, where the team ran structured customer discovery, refined product strategy, and scoped how smart-city technologies could be applied to improve emergency healthcare access.
• Pivoted from a startup path to an open-source initiative to broaden impact and collaboration, making the decision engine and hospital-signal ingestion auditable and extensible outside the founding team.

Stack notes

Full-stack web (TypeScript end to end), background workers for scheduled and event-driven ingestion of hospital and traffic signals, normalized healthcare-data store, HTTP APIs, responsive client UI optimized for one-handed and high-stress use, and an optional on-device passive safety hook for automatic emergency dispatch.