Inspiration
For years, I have watched hospitals struggle to understand perioperative performance because scheduling, readiness, staffing, supplies, clinical milestones, recovery, billing, and collections live in separate systems. Surgemetry began as a way to connect those fragments around the surgical case and turn them into accountable leadership action.
What it does
Surgemetry is a case-centered perioperative governance and accountability layer. The synthetic prototype connects:
- case readiness dependencies and hard blockers
- OR flow, block use, turnover, and after-hours activity
- anesthesia coverage and ASA-unit context
- journey events, cancellations, PACU, discharge, and disruption attribution
- directional expense, reimbursement, denial, payment, and collection signals
- data-quality warnings, governance prompts, and an auditable local event trail
The dashboard shows not only what happened, but what needs review, who the leadership audience is, and what information or action may be missing. It does not replace Epic, Oracle Health/Cerner, finance, supply-chain, billing, or other systems of record. It provides a future vendor-neutral reconciliation and governance layer above them.
How we built it
Codex helped translate specialized perioperative operations knowledge into a working Python prototype. It was used to inspect and evolve the repository, build deterministic synthetic datasets and calculations, refine the Streamlit experience, add tests and documentation, create a clean demo, and run privacy, copyright, and release-safety checks.
The stack includes Python, pandas, NumPy, Streamlit, Altair, Pydantic, pytest, local CSV files, Markdown reporting, a local SHA-256 hash chain, and an optional OpenAI Responses API feature.
All operational and financial metrics remain deterministic. GPT-5.6 Sol has one narrow role: after an explicit click, it converts an already-computed governance signal plus allowlisted aggregate synthetic context into a structured executive-review draft. It does not calculate metrics, set thresholds, change recommendations, retrieve sources, or make decisions. A deterministic fallback is used if the model is unavailable or its output is invalid.
Challenges
The hardest design challenge was resisting the temptation to build another dashboard. A useful perioperative product must connect fragmented dependencies, preserve true causes, distinguish hard blockers from averages, assign accountable review, and link operational events to financial consequences without creating false precision.
We also had to define honest boundaries. The prototype has no live hospital connectors, no PHI, no validated clinical or financial conclusions, no production security controls, and no claim of HIPAA compliance. Epic, Oracle Health/Cerner, AdaptX, warehouses, HL7v2, FHIR, SMART, SFTP, and vendor APIs are prospective integration paths, not implemented capabilities.
Accomplishments
- Built a case-centered vertical slice across seven synthetic datasets
- Added readiness, journey, PACU, cancellation, disruption, and directional financial views
- Preserved true elapsed gaps separately from capped turnover
- Added reconciliation labels and double-counting protections
- Created nine Streamlit tabs with case-level drill-downs
- Added regenerable reporting and a local tamper-evident audit demonstration
- Added an optional, bounded GPT-5.6 executive-note workflow
- Created automated tests and copyright-safe evidence controls
What we learned
The differentiation is not a larger catalog of familiar OR metrics. It is a case-centered governance layer that joins dependency visibility, action ownership, disruption attribution, longitudinal workflow performance, and operational-financial interpretation.
AI is most credible here when it is bounded. Deterministic calculations remain authoritative; the model helps draft executive language from controlled aggregate inputs, with human review and clear labeling.
What's next
Next steps include status history, intervention assignment and escalation, acknowledgment and resolution workflows, deeper cancellation and recovery analysis, stronger financial reconciliation, configurable governance thresholds, controlled pilot file mappings, and then selective batch or near-real-time connectors.
Production use would require authentication, authorization, encryption, retention controls, audit architecture, privacy governance, validation, and an approved HIPAA-ready environment.
Safety: Every case, person, organization, and result in this demo is synthetic. Surgemetry is operational decision support, not clinical guidance, and is not production-ready or HIPAA-compliant.
Log in or sign up for Devpost to join the conversation.