AI Automation
Clawleaf: Intake, Triage & Routing
Faxes, inbox mail and stubborn hospital portals read, classified and routed to the right team.
DisciplinesAI Automation
Our roleDesign, build, release
Technologies
Python
PyTorchVision modelsInference at edge
PythonPyTorchVision modelsInference at edge
Clawleaf
The challenge
What was in the way.
Inbound work arrived as faxes and shared-inbox mail with no owner, and the data needed to act on it was locked behind hospital portals with interfaces built for humans, not machines.
Disciplines in scope
Hardware—
Firmware / IoT—
AI / MLIn scope
Web & mobile—
How we built it.
Documents are classified on sender, recipient and topic together rather than on any single signal, and every routed item carries a link back to the original so staff can verify in one click. The portal bots are written against the rendered interface with retry and drift detection, so a layout change surfaces as an alert instead of silent data loss.
01Technical discovery
02Architecture & prototype
03Build & iterate
04Deploy & hand over
Fax Intake & Routing Bot — finds unlinked incoming faxes, identifies sender, recipient and topic, and posts each to the right team channel
Inbox Triage & Routing Bot — reads a shared inbox, checks each message's zone, and routes it to the right staff channel
Complex-Portal Data Bot — extracts and summarizes data from a hospital portal with a difficult interface
Portal Data Extractor — pulls structured data out of a clinician web portal for downstream use
01Shipped
Delivered to production and still maintained by the team that built it.02Tested
One test matrix across hardware, firmware and interface — no gaps between layers.03Documented
Runbooks, monitoring and escalation handed over with the system.04Scaled
Grew with usage without re-architecture or a second rebuild.