A digital colleague · not a chatbot
Hire a colleague.Not anothervoice AItool.
BAA across the stack·Zero PHI in reasoning·Audit customers can query
iBridge builds voice colleagues for US healthcare operations. Trained by voice, bounded by authority, engineered to refuse what they haven't been taught. The same discipline runs every agent we ship — for clinics, for RCM, for the workflows your team is drowning in.
Live in production · Specialty Care Clinics · BAA-covered pipeline · 25 years shipping
Engineering posture
↗ See the architecturePHI evaporates
here.
Patient information lives only at the boundary, only for the duration of one interaction. The core reasons on typed proofs. Persistence sees no PHI. Architecturally enforced. Customer-auditable.
See it · demo.smarna.org/infusion · 60 seconds, no PHI→Knowledge architecture
↗ See the four knowledge layersEvery Nora gets smarter from every other Nora.
Verified-public discoveries propagate across tenants. Private workflows and patient data stay isolated to the tenant that owns them. The opposite of consumer AI silos. Architecturally enforced.
Flows across tenants
Payer IVR maps. Refusal patterns. Escalation rules that worked. Verified by a human in the originating tenant. Cross-validated. Audit-logged. Reversible.
Stays with the tenant
Your scope policy. Your workflows. Your authorization rules. Your specialty mix. Patient data heard during one call. Cross-tenant access is architecturally impossible.
Why it matters
Consumer AI shares everything. Walled-garden vendors share nothing. We do neither. Every colleague compounds — without ever sharing what isn't supposed to be shared.
Compliance posture
Named categories · vendors on request
BAA
across the stack.
Signed. Filed. Auditable.
Every category of subprocessor in our pipeline operates under an executed Business Associate Agreement. Not something we're working toward. Something we can send to your compliance team this week.
Security questionnaire response · 5 business days. Not 3 weeks. Ask us.
Pilots in progress
Live
Specialty Care Clinics · infusion insurance verification · production
Pilot 01
American Specialty Clinics · multispecialty · Texas
Pilot 02
American Infusion Care · infusion · prior auth
Built on
Meet the colleagues
Named · scoped · audit-logged
They have names.
They have authority charts. They have memory that grows.
Nora
LiveInsurance verification · payer calls
Nora calls Aetna, BCBS, United. Captures visit limits, prior auth status, site-of-service requirements, copay. Every field she captures has a confidence score. Below 70%, she flags for human review instead of guessing.
Live at Specialty Care Clinics · production
Watch Nora call your payer · 60 seconds→Nora at a glance
Maya
Scoped · ready to deployPatient access · scheduling · triage
Maya answers inbound calls for scheduling, refills, intake, after-hours triage. Hands off complex cases to staff with full context. Refuses anything outside the autonomy policy her clinic owner has set. Same architectural discipline as Nora. Same training authority model. New workflow, new customer.
Maya at a glance
New colleagues onboard the way new staff do: assigned to a workflow, given a phone number, paired with a training authority and one or two peers, then taught by voice. Same architectural discipline for every colleague we ship — whether we deploy them as Nora, as Maya, or under your brand.
How Smarna learns
Training authority · not a permission set
No prompt editor.
No rule forms.
No admin panel.
She's taught the way humans are taught. A senior peer dials her number, teaches in conversation, and she asks clarifiers when she's unsure. Nothing changes until the right human approves.
Voice-recognized peers·Reasoning shown·Routing announced
01 · TEACH
A peer teaches her by voice.
"Hi Lokesh, what would you like me to learn today?"
The peer is recognized by voice fingerprint. The colleague loads the full authority chart for that workflow before saying a word. Teaching arrives as conversation, not configuration.
02 · REASON
She tests the teaching against what she knows.
"That contradicts the CMS guidance from last quarter — let me check."
The colleague reasons across four knowledge layers. Contradictions surface. Clarifiers are asked. Layer B sources are consulted live and logged.
03 · ROUTE
She announces what she will do.
"Since you're a peer and this is in-scope, I'll route this to Jennifer for verification before it changes my behavior on calls."
Authority and scope are evaluated. In-scope teachings go to verification. Out-of-scope changes create an escalation flag with the full reasoning chain attached.
What she isn't
Every other voice AI vendor
Prompt editors. Rule-form builders. Config JSON. Admin panels. Clinic staff typing conditions into fields they don't understand.
iBridge
A conversation with authorized humans. She refuses what she hasn't been taught. Every teaching moment is audit-logged with the human who taught it.
Solutions
03 / Workflow
Where we are goingVoice is one interface.
Documents and workflows are next.
Most operational work in healthcare isn't a phone call. It's a fax that arrives and waits to be keyed in. A denial that needs categorizing. A schedule that needs preparing. The same architectural discipline as our voice agents, applied to the documents and workflows your team is drowning in.
Scope a workflow engagement →01 / Clinics
For 30 to 80 provider multispecialty groups.
Maya handles the routine 70 percent — scheduling, no-show recovery, refill triage, after-hours routing, intake.
Explore clinic agents →02 / RCM
For RCM teams running prior auth at scale.
Nora calls payers for benefits, prior auth status, claim follow-up, denial capture. Days to hours.
Explore RCM agents →Live in production
Built with Specialty Care Clinics
Meet Smarna.
A digital colleague we shipped.
Smarna runs infusion insurance verification at Specialty Care Clinics — calling Aetna, BCBS, United, capturing visit limits, prior auth status, copay, and site-of-service requirements with confidence scores on every field. Live. In production. Today.
She's one example of what we build. The same architectural discipline — PHI-zero core, hash-chained audit, voice- trained learning, refusal logic — powers every agent we ship for clients, whether we deploy as Smarna or build under your brand.
Boundary
PHI lives only here. Sealed. Destroyed per call.
Core
Reasons on typed proofs. Never on PHI.
Audit
Hash-chained. Customer-queryable. Day one.
Proof in production
Customer-queryable · from day one
We don't demo screenshots.
We ship the surfaces that produce them.
100%
PHI destruction pass rate
Every call ends with a cryptographic proof that the in-memory PHI buffer was destroyed. A proof passes when zero violations are detected across scanned turns. The proof hash is stored so any single call can be independently verified after the fact.
61
Calls made
rolling
28
Recordings
rolling
233
Fields captured
rolling
58
PHI proofs
rolling
Recent PHI destruction proofs
Auto-refreshing
Sample structure · your tenant shows live production values
89.4%
Average capture confidence
across 233 captures
44.4%
Capture completeness
8 of 18 calls captured patient responsibility
71
Total verifications
across the platform's lifetime
Confidence distribution
233 captures
Capture completeness is 44%. That's honest: Nora captures every field she's confident about, and refuses the ones she isn't. Some fields require a human today. We publish this number because the vendors who claim 95% completeness are the vendors whose agents are guessing.
Sample structure · your tenant shows live production values
01 · Live call ledger
Every inbound and outbound call, newest first. Timestamp, direction, status, duration, destination, per-call cost. Click any row for full transcript replay with captured fields and audit chain.
02 · Training inbox
Provisional teachings captured during voice-training calls, waiting for the training authority to approve, correct, or reject. Each teaching carries a confidence score and reasoning chain.
03 · Recordings & transcripts
Full audio recordings and identifier-redacted transcripts, searchable by VOB, patient, drug, or payer. Retention window customer-configurable. BAA-covered storage.
Also shipped · live activity ticker
Synthetic preview · customer feeds are scrubbed
Live · activity
tail -f /var/log/colleague.activity
connecting...
synthetic preview · live customer feeds are scrubbed by architecture
Every iBridge customer gets these surfaces from day one. Procurement teams audit us with them. Operations teams supervise the colleague with them. Engineering teams confirm the architecture matches the marketing.
Watch a call resolve
Patient calls.
Maya listens.
Your system gets a proof.
This is a representative transcript of Maya resolving a real-world inbound scheduling call. Watch her work, then watch the structured output your downstream system receives. No raw transcript stored. No PHI persisted.
Compliance posture
BAA-ready
Signed Business Associate Agreements with every customer and every subprocessor in the chain.
US-only
All inference, storage, and telephony deployed in US-only AWS regions. No cross-border data movement.
Audit-everywhere
Every prompt, decision, knowledge consultation, and state change captured in a hash-chained log. Customer-queryable.
Reading
Field notes
we're publishing.
Articles for procurement, RCM directors, and practice administrators evaluating voice AI for healthcare.
Read all insights
Kalpesh Upadhyay
Founder · 25 years shipping production software in regulated industries