Problem framing speaker notes
Keep this page on your own screen. The presentation page now contains only the material the interviewers should see.
Pillar 01
Connect and Understand
Build one reliable, current view from fragmented clinical information.
One-line impact
Clinicians spend time reconstructing the patient before they can reason about care.
“Patient information is not missing; it is fragmented. The EHR, claims, labs, pharmacy records, documents and messages each hold part of the truth. Before Praxis can reason, it has to reconcile those sources into a current patient state and preserve where every fact came from.”
EHR data
Problem in simple terms
Medications, diagnoses, labs, notes and encounters live in separate parts of the chart.
Example
A specialist discontinues a medication in their note, but the primary-care medication list still shows it as active.
Impact
Reviewing one section alone creates an incomplete patient story.
Claims, labs and pharmacy
Problem in simple terms
These sources use different codes and arrive at different speeds.
Example
A patient fills a medication at an outside pharmacy, but the claim arrives weeks later while the EHR still shows the therapy as overdue.
Impact
New risks can appear late, out of order or without enough clinical context.
Documents and notes
Problem in simple terms
Important facts are often buried inside long, unstructured records.
Example
A discharge PDF mentions an acute kidney injury and a medication change, but neither fact appears in structured fields.
Impact
Manual review is slow and clinically important evidence can be missed.
Messages and inbox events
Problem in simple terms
Symptoms, refill requests and questions change the patient between visits.
Example
After a dose increase, a patient reports dizziness through the portal, but the structured chart still looks unchanged.
Impact
The care plan can become stale before the structured chart changes.
History, freshness and provenance
Problem in simple terms
Facts can be old, duplicated, contradicted or superseded by later evidence.
Example
Two medication lists disagree; Praxis needs to know which was reconciled most recently and who verified it.
Impact
Praxis must know what is true, when it was true and where it came from.
Pillar 02
Reason and Prioritize
Decide what matters for one patient—and who matters across the panel.
One-line impact
Attention follows the schedule and the loudest alerts, not always the greatest need.
“Once the data is connected, Praxis still has to decide what deserves attention. That means reasoning across conditions and time for one patient, then comparing urgency across an entire panel. The system cannot use the most expensive reasoning on every patient, so prioritization itself becomes a core clinical and engineering problem.”
Patient-level reasoning
Problem in simple terms
A recommendation must account for the whole patient, not one isolated condition.
Example
A diabetes agent suggests intensifying treatment, but recurrent hypoglycemia and kidney disease make that unsafe.
Impact
Condition-specific advice can conflict with comorbidities, medications or goals.
Care gaps and changing risk
Problem in simple terms
The useful question is what changed and whether it needs action now.
Example
An overdue eye exam is important, but a rapidly rising creatinine level requires attention first.
Impact
Static reports create noise while meaningful deterioration remains hidden.
Panel prioritization
Problem in simple terms
A clinician may be accountable for roughly 1,200 patients at once.
Example
Two hundred patients have open care gaps, but only twelve show recent deterioration that may need action this week.
Impact
Deep reasoning for everyone is too expensive; simple sorting is too shallow.
Cross-time and condition reasoning
Problem in simple terms
Clinical meaning depends on sequence, trends, conflicts and later evidence.
Example
One high blood-pressure reading during acute pain means something different from a steady rise across six weeks.
Impact
A correct fact used at the wrong time can still produce the wrong recommendation.
Practice patterns
Problem in simple terms
Improvement requires comparing decisions and outcomes across many patients.
Example
A clinician repeatedly delays statin discussions for eligible diabetes patients—a pattern invisible in any single visit.
Impact
Systematic gaps cannot be seen from individual encounters alone.
Pillar 03
Coordinate and Act
Turn a useful insight into owned, trackable and completed work.
One-line impact
A recommendation that never reaches execution does not improve care.
“The next problem is execution. Praxis should not stop after producing a recommendation. It needs to turn that recommendation into the right workflow, assign responsibility, request approval when required and track the work until it is completed or deliberately dismissed.”
Inbox work
Problem in simple terms
Messages, results, refills and staff questions need different response paths.
Example
A routine refill can be drafted, while a message mentioning chest pain must be escalated immediately rather than answered normally.
Impact
Clinicians repeatedly reconstruct context before making small decisions.
Follow-up and outreach
Problem in simple terms
A recommendation must become a task with an owner, timing and next step.
Example
Praxis detects an abnormal potassium result, but nobody is assigned to contact the patient or repeat the test.
Impact
Patients fall between identifying a need and actually contacting them.
Orders and clinical actions
Problem in simple terms
Low-risk work can be prepared; higher-risk actions require approval.
Example
Praxis can prepare a routine lab order, but changing an anticoagulant dose must wait for clinician approval.
Impact
Unclear autonomy boundaries make the system either unsafe or not useful.
Care transitions
Problem in simple terms
Admissions, discharges and outside care create time-sensitive follow-through.
Example
A hospital discharge says to stop lisinopril, but the outpatient EHR still lists it as active and no follow-up is scheduled.
Impact
Delay increases readmission, duplication and medication-error risk.
Completion tracking
Problem in simple terms
Praxis must know if work was approved, attempted, completed or dismissed.
Example
An outreach task is created, but the patient is never reached; without status tracking, the system may assume the loop is closed.
Impact
Otherwise, recommendations become another inbox rather than reducing work.
Pillar 04
Trust and Improve
Make asynchronous clinical work explainable, bounded and measurable.
One-line impact
An unwatched error can repeat across a panel before anyone notices.
“Finally, async work changes the safety model. The clinician did not watch the reasoning happen, and one failure can repeat across hundreds of patients. Praxis therefore has to show its evidence, reconcile conflicting outputs, recheck freshness before action and measure quality before clinicians have to discover problems themselves.”
Evidence and citations
Problem in simple terms
Every recommendation must connect to patient facts and clinical guidance.
Example
Praxis recommends changing therapy but cannot show the eGFR value or guideline passage that supports the decision.
Impact
Clinicians cannot verify a conclusion without inspectable evidence.
Validation and reconciliation
Problem in simple terms
Agent outputs can conflict, duplicate work or use unsupported facts.
Example
The diabetes agent recommends metformin while the renal agent flags the patient’s kidney function as a contraindication.
Impact
A confident final answer can hide unresolved disagreement.
Freshness and accountability
Problem in simple terms
Patient state can change after analysis, and actions need a human owner.
Example
A recommendation created from Monday’s labs becomes unsafe after the patient is admitted on Wednesday.
Impact
A previously reasonable action can become unsafe before execution.
Evaluation and monitoring
Problem in simple terms
Quality includes omissions, ranking, safety, latency and cost—not fluency alone.
Example
Inbox drafts sound polished, but the system misses a small number of urgent messages or ranks them too low.
Impact
The team otherwise cannot tell if a new version is better or merely different.
Feedback, memory and alert burden
Problem in simple terms
Praxis should learn preferences without learning unsafe clinical habits.
Example
A clinician dismisses one reminder for a documented reason; Praxis should remember the context without suppressing that reminder for every future patient.
Impact
Bad memory creates bias; no memory creates repetitive, low-value alerts.
Closing transition
“These are connected problems, not four separate products. Better data improves reasoning, better reasoning creates better work, and safe execution creates the feedback loop that improves the system.”