Custom AI for Medical Supply Inventory Management

Outpatient clinics often rely on the inventory modules within their EHR, like those in eClinicalWorks or Allscripts, or a standalone tool like Hybrent. These systems track consumption well but are reactive. They operate on static PAR (Periodic Automatic Replenishment) levels, triggering a reorder only after stock falls below a fixed number. This model completely ignores future demand, which is the most important signal in a clinical setting.

Consider this scenario: it is late September, and the clinic just opened up 50 new appointment slots for flu shots next week. A nurse practitioner also scheduled 15 non-routine dermatological procedures. The PAR-level system sees that current syringe and lidocaine stock is adequate. It has no connection to the appointment schedule, so it cannot know that demand will triple next week. The result is a stockout on Tuesday morning, forcing staff to reschedule patients and place an expensive emergency order.

This is not a feature gap; it is an architectural limitation. EHR inventory modules are designed as ledgers, not predictive engines. They are built to record transactions that have already happened. Connecting the calendar (future events) to the stockroom (current state) requires a separate layer of logic that these monolithic systems were never designed to support. You cannot add a custom forecasting model to your EHR's baked-in inventory tool.

The first step is a data and workflow audit. Syntora would map your clinic's patient scheduling process, key appointment types, and the specific supplies required for each. We would analyze how to access schedule data from your EHR, either through a direct API or a secure database export. This discovery phase produces a clear technical plan and a fixed-price proposal before any code is written.

The technical approach involves a Python service deployed on AWS Lambda that runs on a schedule, typically every 4 hours. This service pulls appointment data for the next 14 days. It uses a mapping logic, potentially enhanced by a call to the Claude API for parsing unstructured appointment notes, to translate visit types into a bill of materials. The system then compares this projected demand against current inventory levels queried from your existing system to generate a recommended purchase order stored in a Supabase database.

The delivered system is a secure, HIPAA-compliant automation that requires minimal staff interaction. A clinic manager receives a daily email with a link to a simple web interface showing pending purchase orders. They can review, approve, or edit any order with a single click. The approved order is then automatically submitted to the supplier's portal. This provides a human review gate while automating over 95% of the manual calculation and data entry.