A discharge summary tool built on Gemini 2.5 Flash that reads hospital PDFs and instantly surfaces the three things every care transition team needs to know.
Hospital discharge PDFs average 30–140 pages. The team coordinating the next step in care — rehab, home health, skilled nursing — needs the same three things every time.
A clinician or coordinator would spend 15–25 minutes reading through a full discharge summary just to extract the hospital course, therapy notes, and prior function level — information buried across multiple sections.
What brought them in? What did rehab find? What were they doing before the hospital? These questions never change — but finding the answers in an unstructured PDF always took the same manual effort.
The original approach relied on emailing PDFs and waiting for a reply. When that process broke down — wrong labels, API issues, quota limits — the whole workflow stalled.
A single-page web app that reads any hospital discharge PDF and returns a structured clinical summary — with no login, no backend, and no data stored anywhere.
Drag and drop any discharge summary PDF. The file is read entirely in the browser — it never touches a server.
The PDF is base64-encoded and sent directly to Gemini 2.5 Flash via the API. The model reads the entire document — not just extracted text — and answers three structured clinical questions in a single pass.
Gemini 2.5 Flash · Google AI APIHospital course summary, rehab therapy findings, and prior level of function are returned as structured sections — ready to read, copy, or download as a .txt file.
After the initial summary, the user can ask any additional question about the document — medication reconciliation, specific diagnoses, discharge instructions — without re-uploading or losing context.
Every document is analyzed against three consistent clinical questions — the same ones a care coordinator would ask manually.
The first version of this tool lived inside Google Apps Script as an email-based workflow. Send a PDF to a specific Gmail address, get a reply. It worked, but the fragility of email triggers, API quota limits, and label management made it unreliable in practice.
The shift to a browser-based web app solved every infrastructure problem in one move. No email routing. No trigger management. No server. The user controls the workflow entirely — drop a file, get an answer.
What took an entire clinical session to read and summarize manually now takes less time than it takes to open the document. That's the compounding value of matching the right AI model to a narrow, well-defined task.
"The same three questions, answered the same way, for every patient — that's exactly the kind of narrow, repetitive task that AI handles best. The hard part was just building the right wrapper around it."
The follow-up question feature — added after the initial build — turned out to be more valuable than expected. Rather than re-reading the document for a specific detail, the user can ask naturally: "What were the discharge medications?" or "Was there any mention of wound care?" The model already has the full document in context.
Most tools that process sensitive documents run your files through a backend server — a database somewhere stores what you uploaded, when, and what came back. That's a liability for clinical workflows where patient information is involved.
This tool has no backend. When you enter a Gemini API key, it's saved only to your own browser's local storage — never transmitted anywhere except directly to Google's API when you click Analyze. Your PDF goes from your browser straight to Gemini and nowhere else. No intermediate server. No database. No logs.
The tradeoff is one extra setup step: getting a free API key from Google AI Studio. Once it's saved in your browser, you never have to enter it again. And because the key is yours, you control the usage, the billing, and the data — not us.
"No backend means no breach surface. Patient documents never touch a third-party server — just your browser and Google's API, with a key only you hold."
Get a free Gemini API key from Google AI Studio, paste it in once, and upload any hospital discharge PDF. Nothing is stored — everything runs in your browser.
Open the PDF Analyzer →