Six product outcomes. Ten seconds to orient.
If you only scan one section before the detailed capability rows, scan this one. These six outcomes define what PulseEdge is designed to protect.
Works without internet
Registration, capture, and drafting continue locally so the workflow does not stall on signal.
AI stays within boundaries
Environment choice sets AI scope, data visibility, and governance posture together.
Every action is auditable
Original capture is preserved, working edits are separated, and provenance remains reviewable.
No temporary identities
Permanent encounter numbers are created on-device at the point of care, not after sync.
Documents are verifiable
OCR, AI extraction, indexing, and links are split into visible stages with separate reruns.
AI shows what it does not know
Draft Builder marks missing required fields instead of hiding uncertainty behind a polished note.
The six sections below map directly to real product states already visible in the current UI: environment entry, clinician login, patient context, document processing, and Draft Builder.
AI scope, data visibility, and governance move together.
Environment choice sets AI scope, data visibility, and audit posture together.
PulseEdge treats working environment as a first-class operating choice. On entry, the clinician selects the setting, and AI scope, retrieval boundaries, and governance posture all switch with it.
This is not a simple permission flag. Visibility rules, cloud fallback eligibility, audit behaviour, and retention posture are derived from the same environment tier.
- Locked — enrollment-only visibility, restricted AI, forensic audit controls. Example: psychiatry, forensic.
- Restricted — unit-scoped visibility, local-only AI, stricter governance. Example: specialist clinics.
- Standard — full workflow visibility, hybrid AI. Example: outpatient, outreach.
Clinical identifiers that never wait for network.
Permanent IDs are created at the point of care, not after sync.
In outreach settings, registration cannot wait for connectivity, and temporary IDs create downstream risk. PulseEdge generates deterministic, collision-resistant encounter numbers on-device from deployment-scoped ranges.
The local number is permanent. HIS linkage is appended when sync completes, without rewriting the encounter or breaking references.
- Numbering scheme: PE-YYYY-NNNNN, allocated per deployment range
- Zero collision across devices in the same outreach deployment
- Local number persists as permanent reference after HIS reconciliation
- Works identically in Tier 1, Tier 2, and Tier 3 deployment modes
Append-only source. Editable working copy.
The original utterance is preserved even when the draft changes.
Clinical AI should never overwrite its own source. PulseEdge separates raw capture from the editable working copy so the original utterance remains preserved.
The source transcript is append-only. Clinicians can refine the working transcript before extraction, and structured fields stay linked back to the captured source.
- Source — append-only, persisted as-is, never modified after commit
- Working — editable by clinician, diff against source retained
- Extraction — structured fields derived from working, linked to source range
HTN context · Patient-aware range active
Flags that know who the patient is.
Flags evaluate values against patient context, not generic thresholds.
Threshold colouring without context is a shallow clinical signal. PulseEdge evaluates each value against the patient's active conditions, pathway, medication context, and recent trend.
The resulting flag is computed at entry time and stored with the record, so the clinical reasoning can be reviewed later.
- Active conditions (HTN, DM, respiratory, and related context)
- Treat-to-target pathway assignments
- Prior trend from recent entries of the same field
- Medication-contingent thresholds
Four-stage pipeline. Every stage visible.
OCR, AI, indexing, and links are separate visible stages.
Document processing is split into visible, rerunnable stages rather than one opaque "processing" step. Clinicians can see which stage succeeded and rerun only that stage if needed.
OCR text, structured extraction, retrieval indexing, and cross-record links each keep their own provenance instead of collapsing into a single black box.
- Raw OCR text is stored append-only, never rewritten
- Finalized documents become read-only at the record level
- Re-run operations create new extraction runs, not in-place edits
- Document provenance is retained across re-runs
Fever and cough for 3 days. Dry cough.
History of Present Illness
— incomplete —
Drafts flag what's missing.
Draft Builder marks required gaps instead of hiding uncertainty.
Partial drafts should not masquerade as complete notes. PulseEdge validates every generated draft against the template's required fields and marks missing sections explicitly.
Recover Missing Details targets only the unresolved fields, and signing stays gated until completeness requirements are met.
- Primary Care Note · chief complaint, HPI, exam, plan
- Pain Assessment · scoring, location, character, onset
- Diabetes Monitoring · glucose, HbA1c, foot check, renal risk
- Respiratory / Asthma · peak flow, SpO2, wheeze, trigger review
- Growth · paediatric weight, height, BMI, head circumference
- Custom JSON schema charts, versioned with diff preview
Under the hood. What the workflow depends on.
The capabilities above are product-facing. The stack below is the engineering substrate that makes them operationally credible.
◆ Data
- SQLCipher local database Encrypted at rest · device-local key never transmitted
- Drift ORM · schema v50 37 tables · append-only audit events
- sqlite-vec embeddings Local semantic index · ONNX all-MiniLM-L6-v2
- Deterministic migrations Versioned schema · migration invariant preserved
◆ Sync
- Three-tier architecture Outreach · Clinic server · Hospital sync
- PowerSync + Postgres backend Device-to-device <5s on LAN · proven Apr 2026
- Device-to-device WiFi relay mDNS discovery · ECDH session keys · no cloud required
- HIS façade integration TakeCare submission · provenance · rollback per sync job
◆ AI
- On-device inference (llama.cpp) Primary path · no cloud dependency
- Groq cloud fallback llama-3.3-70b · hybrid mode only
- RAG over signed notes Local embeddings · embeddings never sync
- Hallucination guards [INFERRED] tagging · structured prompt architecture
Choose the next conversation: architecture or deployment.
PulseEdge can be discussed as a technical system, a workflow product, or a pilot path. Start with the briefing that matches your role.
Request technical briefing
Review architecture, data posture, sync model, AI boundaries, audit design, and HIS integration detail.
Discuss pilot deployment
Review workflow fit, rollout scope, governance posture, and how PulseEdge can extend existing hospital operations safely.
Interested in the broader opportunity?
PulseEdge was built to solve an infrastructure-level healthcare problem, not just a documentation problem.