Presenter view

Product direction speaker notes

Explain the product before explaining the architecture: one continuous capability loop supporting three workflows.

Opening cue

“I’m designing Praxis to evolve from a single-patient encounter assistant into an asynchronous clinical intelligence and coordination layer.”

Simple definition

“Praxis continuously watches patients across time, identifies meaningful changes, prioritizes who needs attention, prepares the appropriate work and routes it to the right person—with evidence and human approval.”

• Not another dashboard

• Not only a chatbot

• Not an autonomous clinician

Capability 01

Understand continuously

Maintain a current patient picture without waiting for the clinician to open the chart.

Product promise

The clinician should not need to reconstruct the patient story before every decision.

Points to cover

Combine EHR, claims, labs, pharmacy, documents and messages

Understand what changed across time

Preserve the source and freshness of every fact

Identify missing or conflicting information

Simple example

A patient reports dizziness after a medication change. Praxis connects the message with the recent prescription and lab history, then updates the patient’s current state.

Capability 02

Decide what matters now

Determine what requires attention for one patient and across the entire panel.

Product promise

Clinician attention should follow clinical need—not the schedule, the loudest alert or a static list.

Points to cover

Detect meaningful changes, risks and care gaps

Reason across conditions and time

Compare urgency across patients

Explain why one patient is ranked above another

Use deeper reasoning only when necessary

Simple example

Two hundred patients have open care gaps, but twelve show recent deterioration. Praxis moves those twelve to the top and explains why.

Capability 03

Coordinate the work

Convert a useful clinical insight into owned, trackable and completed work.

Product promise

A recommendation is not finished until the work is completed or deliberately dismissed.

Points to cover

Choose the correct workflow

Prepare messages, orders or follow-up tasks

Route work to the appropriate team member

Request approval based on risk

Track completion, failure and escalation

Simple example

Praxis detects an abnormal potassium result, prepares the follow-up plan, assigns the task and escalates it if nobody responds.

Capability 04

Learn and improve safely

Improve from clinician feedback and outcomes without silently learning unsafe behavior.

Product promise

Praxis should become more useful over time without becoming less predictable or less safe.

Points to cover

Show evidence and citations

Record approvals, edits and rejections

Measure recommendation and ranking quality

Identify practice-level patterns

Learn workflow preferences separately from clinical truth

Monitor safety, latency and cost

Simple example

A clinician repeatedly edits the tone of outreach messages. Praxis can remember that preference, but it must not treat a rejected clinical recommendation as a new medical rule.

Connect to the assignment

One product, three workflows

“What should I focus on this week?”

Understand the panel → prioritize patients → explain the ranking

A focused, evidence-backed worklist

“Draft responses to my inbox.”

Understand the message and patient → prepare the response → request approval

Safe drafts ready for clinician review

“How can I improve my diabetes practice?”

Analyze decisions and outcomes → identify patterns → recommend improvements

Practice-level learning grounded in patient evidence

Closing transition

“These are not three separate products. They share one clinical intelligence layer and one patient-state model. The architecture I’ll show next is designed around that shared core.”