Clinical workflows
We learned how outpatient practices register patients, deliver care, coordinate work, and operate.
September to December 2026
From POC
to a commercial EHR MVP
Turning four months of product learning into a healthcare-grade platform clinics can pilot.
Invite outpatient pilot clinics by year-end, then commercially offer Summit as a secure, HIPAA-compliant EHR built on strong healthcare standards.
Use arrow keys to present • N for notes • F for full screen
User brief; RESEARCH-PLAN.md; A01 through A05 settled handbook pages.
The journey so far
Four months well spent proving what Summit should become.
Business needs
We tested the product thesis against real clinic needs and discovered what matters in daily use.
The whole product
We built and hosted a modern outpatient EHR experience across staff, patient, and clinic workflows.
The next build
We assembled the team and learned the healthcare standards, architecture, and AI patterns we need.
The result
We are not starting over. We are carrying proven workflows, reference UI, components, product knowledge, and a prepared team into the commercial build.
references/summit-alpha/context/01-current-state.md; references/summit-alpha/v2/findings/D-product-inventory.md; artifacts/src/application-layer.html; artifacts/src/ai-layer.html.
Where Summit is today
A hosted POC that proves the modern outpatient EHR experience.
The product is usable today. The foundation is not yet commercially ready.
Patients
BookRegisterComplete intakeJoin visitMessage and pay
Clinical staff
ScheduleChartRun visitsPrescribe and referFollow up
Owners
OnboardManage teamConfigure clinicReview operationsGrow
Keep it hosted: clinics can continue internal workflows, registration, digital intake, eligibility, and the capabilities already in use.
Lock feature development: support bugs and small operating improvements while the team focuses new product work in the commercial repository.
Carry the learning forward: reuse reference UI, useful components, workflow design, product knowledge, and the team that built it.
references/summit-alpha/v2/findings/D-product-inventory.md sections 1 through 3; references/summit-alpha/context/01-current-state.md.
Current architecture
The POC stack optimized for learning and speed.
INTERFACE
Next.js on Vercel
Staff, patient, public, admin
THE CURRENT HUB
Convex
- Database
- Business logic
- Authentication
- Realtime
- Files
- Scheduled jobs
AI AND MEDIA
Gemini, OpenAI, Deepgram, LiveKit
Multiple paths and runtimes
CLINIC VENDORS
Stripe, Telnyx, DoseSpot, Stedi, Google
Payments, communications, prescribing, eligibility
●
Why it worked: one lightweight stack let a small team prototype a broad product quickly.
●
Why it stops here: data, logic, identity, files, and jobs are tightly coupled around Convex.
●
What commercial needs: healthcare standards, stronger isolation, governed AI, reliability, and compliance evidence.
references/summit-alpha/context/01-current-state.md, verified 2026-08-26; references/summit-alpha/v2/findings/D-product-inventory.md.
Why we are changing
This is a product refoundation, not a server move.
The product knowledge
Proven clinic workflows, useful interface components, templates, and hard-won operating lessons.
The foundation
Convex as the record, custom auth, mirrored clinical tables, ad hoc background work, and fragmented AI access.
The commercial controls
FHIR standards, AWS isolation, audit evidence, governed tools, continuous QA, recovery, and cutover discipline.
Freeze the POC. Build the commercial MVP in a new repository. Reuse the learning, not the limitations.
decisions/AWS-D-022-no-double-run.md; decisions/AWS-D-030-the-milestone-ladder.md; artifacts/src/stack-and-repo.html.
The migration method
Use Medplum for the record. Build Summit for the experience.
20 capabilitiesAdopt
Use Medplum's healthcare model and discard our duplicate implementation.
- Clinical chart
- Identity and access
- Scheduling core
- Tasks and messaging
13 capabilitiesEnhance
Keep Medplum's record underneath and add the Summit product on top.
- Intake designer
- Visit workspace
- Patient portal
- Prescribing and fax
19 capabilitiesBuild
Create what a clinical platform does not provide and what makes Summit distinct.
- AI harness
- Public tokens
- Payments
- Analytics and notifications
Standing rule
Research the Medplum way first. Any deviation needs a reason before it is built.
decisions/AWS-D-002-medplum-first-principle.md; decisions/AWS-D-019-medplum-way-first.md; decisions/AWS-D-023-medplum-shape-our-transport.md; artifacts/src/application-layer.html section 6.
The commercial architecture
The product gains two healthcare-grade cores.
Healthcare backend + AI brain
inside one controlled AWS platform
PatientsClinic teamsPractice owners
↓
SUMMIT EXPERIENCE
Web application
Scheduling, charting, intake, telehealth, portal, operations
Next.js on Vercel
↓
AWSIsolated Summit platform
APPLICATION LAYER
Summit API and workflow services
One controlled path for product logic, vendors, events, payments, and AI
01
HEALTHCARE BACKEND
Medplum
FHIR clinical record, identity, access policies, audit history
PatientsChartsVisitsOrdersTasks
GOVERNED MCP TOOLS
read · draft · propose
scope, audit, human approval
02
AI BRAIN
Two harnesses
Core agents for product workflows. Voice agents for real-time conversations.
Deep AgentsLiveKit AgentsBedrockGPT models
Clinical dataApplication dataFiles and documentsEvents and queuesLogs and auditHealthcare vendors
Medplum keeps the healthcare record standard
The AI brain helps through controlled product tools
AWS provides isolation, scale, recovery, and operational control
artifacts/src/application-layer.html sections 1 through 7; artifacts/src/ai-layer.html sections 3 and 8; artifacts/src/stack-and-repo.html sections 1 and 2; artifacts/src/environments.html section 2.
The Summit AI brain
One AI brain. Two specialized harnesses.
The application stays accountable. The AI layer supplies intelligence through governed tools.
VOICE
LIVEKIT AGENTS
Real-time voice harness
- Voice conversations
- Live visit assistance
- Telehealth media workflows
SUMMIT AI BRAIN
Shared identity, context, tools, traces, and evaluation
MODEL PLANEAmazon Bedrock + GPT family
CORE
DEEP AGENTS
Product agent harness
- Workflow assistants
- Long-running tasks
- Drafts and coordination
APPLICATIONowns the record and workflow
→
MCP TOOLSscope and audit every action
→
AI BRAINreads, drafts, and proposes
→
HUMANreviews clinical commitments
Binding rule: the agent may assist throughout the workflow, but a clinician commits the signed clinical record.
artifacts/src/application-layer.html sections 7 through 9; artifacts/src/ai-layer.html sections 1 through 8; decisions/AWS-D-015-agent-identity-and-injection.md; decisions/AWS-D-029-tool-plane-is-mcp-in-typescript.md; user feedback on the two-harness direction.
Reliability and isolation
Four environments. Two complete AWS deployments.
Staging rehearses production. It never shares production data.
Local
A complete synthetic stack on a developer laptop.
No cloud patient data→
Preview
A temporary Vercel interface for each change.
Points to staging→
Staging
A smaller but complete AWS copy of the backend.
Synthetic data onlyHUMAN GATE→
Production
The regulated system serving real clinics.
Separate network, data, keys, identitySame artifactThe image tested in staging is promoted, never rebuilt.
Separate wallsProduction and staging have no shared database, network, identity, or vendor credentials.
One live SummitThe old product stays live until the manual cutover. There is no double run.
artifacts/src/environments.html sections 1 through 5 and 9; decisions/AWS-D-032-environment-topology.md; decisions/AWS-D-022-no-double-run.md.
What we are building
Twenty-five milestones. Six business outcomes.
1-2
Trust the platform
- AWS environments
- Clinic tenancy
- Staff, roles, access
- Audit foundation
3-5
Build the record
- Patients and chart
- Intake and consent
- Scheduling
- Encounters and tasks
6-7
Transform the visit
- Live transcription
- Clinical assistance
- Draft documentation
- Telehealth
8-13
Close the loop
- Fax and prescribing
- Labs and diagnostics
- Referrals
- Post-visit coordination
14-17
Open the front door
- Payments and eligibility
- Clinic microsite
- Registration kiosk
- Patient portal
18-25
Operate and launch
- Communications and AI
- Exchange and analytics
- Reputation workflows
- Migration and launch
Every milestone ships three things together
the user workflow+its AI tool slice+tests and a recorded demo
artifacts/src/scope-and-order.html sections 1 through 10; decisions/AWS-D-030-the-milestone-ladder.md; decisions/AWS-D-031-qa-is-continuous.md.
What Summit becomes
A lightweight, outpatient-first EHR that covers the clinic day.
Operations
Staff managementPatient managementAppointmentsEligibilityPayments
Clinical workflows
ChartingTelehealthPrescriptionsLab ordersFaxReferrals
Access and communication
Digital intakePatient portalClinic micrositeMessagesNotifications
Assistance and automation
Voice agentsClinical draftsWorkflow assistantsPre-visit supportCoordination
Commercial meaning
A practice receives a lightweight, outpatient-first, AI-native EHR that covers its core healthcare workflows in one accountable system.
artifacts/src/scope-and-order.html milestones 1 through 25; artifacts/src/application-layer.html capability map; references/summit-alpha/v2/findings/D-product-inventory.md.
Four-month execution roadmap
Two months to migrate. Two months to make it clinic-ready.
Year-end targetinvite outpatient pilot clinics
ACT 1Build and migrateComplete the major AWS, Medplum, application, and AI implementation
SEPFoundation
- AWS environments and deployment
- Medplum record and identity
- Application and AI rails
- Patient and clinic spine
OCTMajor build
- Core clinical workflows
- Summit product experience
- Core and voice AI harnesses
- Vendor and event integrations
ACT 2Test and stabilizeTurn the integrated build into a reliable product clinics can use
NOVQuality
- End-to-end clinic-day QA
- Integration and migration testing
- Bug burn-down
- Security and reliability proof
DECPilot readiness
- Polish and stabilization
- Recovery and cutover rehearsal
- QA acceptance
- Pilot onboarding preparation
CONTINUOUS ACROSS ALL FOUR MONTHSQA feedback · compliance evidence · implementation learning · milestone demos
User planning assumption; artifacts/src/scope-and-order.html sections 11 and 12; decisions/AWS-D-031-qa-is-continuous.md; decisions/AWS-D-022-no-double-run.md.
Who is building it
A 10-person build team with clear product ownership.
Designs the architecture, owns technical decisions, and sets product direction.
↔
Owns day-to-day technical planning, coordination, milestones, and delivery rhythm.
4 FULL-TIME
Application engineering
AtikaMariamShahzebHimesh
Application, platform, Medplum migration, integrations, and milestone delivery.
1 TO HIRE
AI engineering
Mohammad Abdullah
Deep Agents core harness, LiveKit voice harness, tools, evaluation, and model operations.
3 TO SOURCE
QA and implementation
Test daily and give rich feedback now, then onboard and support clinics after launch.
FIRST FOUR MONTHSQA learns the product while testing every milestone→AFTER APPROVALThe same team supports implementation and clinic onboarding
User feedback on team roles and planned hiring. Application engineer names are assigned as provided. AI and QA roles remain planned hires.
What success looks like by December
A commercially ready outpatient EHR approved to start pilots.
Core EHR
Staff and patients can complete the agreed outpatient workflows end to end.
Healthcare core
Medplum, FHIR records, identity, access, and audit are operating on AWS.
AI brain
The core and voice harnesses assist through governed tools and human approval.
QA approved
Clinic-day, security, reliability, recovery, migration, and support testing pass.
Pilot operations
Implementation, onboarding, support, and escalation are ready for the first clinics.
GOBegin controlled go-live and pilot onboarding
After QA acceptance and leadership approval of the security, compliance, reliability, and operating evidence.
Architecture creates the conditions for compliance. Completed evidence and leadership review authorize the claim.
artifacts/src/scope-and-order.html sections 11 and 12; RESEARCH-PLAN.md status for A07, A08, A09, and A10; decisions/AWS-D-031-qa-is-continuous.md; user success brief.
The next 16 months
Build. Validate. Scale.
The migration earns the starting line. Pilot learning determines what scales next.
4 monthsSEP TO DEC 2026
Build the commercial MVP
AWS and Medplum migration, core EHR, AI brain, QA, stabilization, and pilot readiness.
6 monthsJAN TO JUN 2027
Pilot and validate
Onboard clinics, study daily use, improve workflows, prove value, and find repeatable product-market fit.
6 monthsJUL TO DEC 2027
Scale product and revenue
Expand revenue cycle, labs, radiology, mobile, freestanding ER support, UI, and core capabilities.
artifacts/src/scope-and-order.html section 10; user future roadmap brief.
The decision now
The POC proved it.
Now build what clinics can buy.
Summit brings the workflows and product knowledge. Medplum and AWS provide the healthcare-grade foundation. The AI brain makes it different. QA and evidence make it commercial.
Lock the pilot scope
Resolve the opening-set versus full-release launch bar.
Start the external clocks
Agreements, terminology, production access, and pilot recruitment.
Name the owners
Workstreams, QA, implementation, compliance evidence, and cutover.
THE OBJECTIVE
Commercially offer Summit to healthcare practices as a secure, HIPAA-compliant EHR built according to strong healthcare standards.
Presentation synthesis grounded in A01 through A05, the 25-milestone ladder, the user brief, and the current-state audit.