Every answer, traced to the page it came from

DocTalk reads the whole chart — decades of PDFs, scanned encounters and lab reports — and answers clinical questions in plain language. Each statement links to the exact page that supports it, so checking it takes one click instead of twenty minutes.

In the current deployment

+1,000: Patient records loaded; Full charts, not summaries
+48,000: Documents indexed; PDF, HTML and plain text
22 min: Documentation time saved per clinician each day; Measured against the previous workflow
100%: Answers carrying a page citation; An unsourced claim is never shown

Provisional figures from the pilot deployment, pending independent audit.

What the platform does

Ask the chart

Natural-language questions across a patient's entire history. The answer carries [S1], [S2] markers that open the exact page: current medication, the last A1c, why a referral was made.

Import and read

A single eCW export can be dozens of encounters in one PDF. DocTalk splits it, extracts text and metadata, and the viewer opens a page range instead of making you scroll a 209-page file.

Consultation to SOAP note

Record the visit, transcribe it in chunks as it happens, and turn the transcript into a structured note with your own prompt. The patient's name is tokenised before the model sees it and restored afterwards.

Scheduling

Day grid by provider, facility and resource. Overlaps are refused at creation — both inside a room and across a provider's whole day.

Standing questions

Save a prompt once and run it against a patient on demand or overnight: discharge summaries, medication reconciliation, checks across eleven configured HEDIS measures.

Coding and payer rules

Extracts CPT and ICD-10 codes from a note or a document range and validates them against a live payer-rule set, with a governance trail for every rule accepted, retired or rejected.

30-day readmission risk

The model extracts the facts — encounters, diagnoses, labs, frailty and utilisation signals — and the score is then computed deterministically. No model opinion enters the number.

The chart never leaves the network

The document engine runs on a private network with no port open to the internet, and the web application is the only thing that talks to it. Identifying data is replaced with tokens before any model call and restored in the answer, so names never reach a third party. Every call is audited against the person who made it.