From Days to Minutes: How AI Transformed Analytical Decision-Making in Healthcare

IMPACT

The less visible gain is harder to quantify but more significant. The system has no knowledge decay. It can surface a combination of biomarkers the doctor has genuinely never encountered, prompt them to look it up, and lead to a referral that would not have happened otherwise. A doctor sees an unfamiliar condition flagged, looks it up, confirms the indicators are consistent, and brings in a specialist. That patient gets care they would otherwise have missed — not because the doctor failed, but because no doctor could be expected to carry seventeen years of missed research in their head.

CHALLENGE

The workflow hadn’t changed in years. A patient arrives carrying a printed lab report. The doctor opens the software, transfers every value by hand, then writes a brief summary based on whatever patterns they recognize. It works — slowly, with real risk of transcription error, and limited entirely by what that one doctor happens to know. The harder problem is that medical knowledge doesn’t stand still. Doctors learn intensively in their early years, then spend the rest of their careers seeing patients — with little time to keep up with new research. Rare but treatable conditions get missed — not from carelessness, but because no one can be expected to carry every diagnostic pattern in their head while running a full patient schedule.

SOLUTION

We built a multi-modal document processing system that begins with classification, because “a document” is not one thing.

Incoming files fall into five distinct categories, each requiring its own extraction strategy:

• Plain images — photos of printed reports, sometimes angled or poorly lit, handled through OCR with noise tolerance built in
• Unstructured PDFs — machine-generated but with no predictable layout; every provider arranges the same tests differently
• Structured PDFs — documents from known providers with fixed, predictable element positions; extracted precisely against a stored template
• OCR PDFs — appearing structured but generated through an unknown external process; treated as images wrapped in a PDF container, because the internal representation cannot be trusted
• Audio — voice notes from the doctor, sometimes supplementing a PDF, sometimes embedded directly