Fragmented documentation
Patient context, encounter notes, prescriptions, and follow-up often live across separate surfaces—or only on paper.
Early product direction · static showcase · v2.1
Chamber Companion is an offline-first documentation and prescription workspace for verified doctors in Bangladesh—designed around the encounter, the source, and the doctor’s final decision. AI suggests. The doctor decides and signs.
The operating reality
It needs one dependable path from patient intake to an attributable, readable record—even when the network disappears.
Patient context, encounter notes, prescriptions, and follow-up often live across separate surfaces—or only on paper.
The core encounter, reference, draft, and document flow must remain available when the device is offline.
A source is not enough. Rules, references, and suggestion inputs require licensing, versions, clinician review, effective dates, and rollback.
Templates and suggestions can reduce repetition, but they cannot bypass patient-specific review or the doctor’s explicit signature.
Three trust zones
The current product direction keeps three surfaces visibly separate. Each zone below is a synthetic container: it shows the shape of the boundary, never a clinical rule, drug, dose, or threshold.
The BMDC-registered doctor writes the assessment, may open AI suggestions, then accepts, edits, ignores, or overrides each one, resolves every check, and signs. Assistants may prepare queue, demographics, and vitals only where permitted.
A patient may opt in to a longitudinal record and grant a named, verified doctor scoped, time-bound, revocable access. The patient is a record-holder and consent controller, not a clinical user.
AI may suggest candidate medicines and investigations, warn about conflict and patient-context risks, and present a candidate dose from doctor-confirmed facts. Every suggestion is source-visible and provisional. It enters the prescription only when the doctor accepts or edits it—and is then independently checked.
Solid cards are doctor actions. Dashed cards are AI or system steps that produce candidates, never prescription items.
Core workflow preview
Use the stepper to explore a synthetic workflow. Step 4 opens the AI suggestion panel. This interface demonstrates information architecture only; it contains no clinical logic.
Every card is a candidate, not a clinical output. Nothing here is part of the prescription until the doctor accepts or edits it. Accepted or edited items then run independent checks before signing.
dose suggestion unavailable · value withheld until formula/source, evaluation, and CSO gates close
Prescription builder preview
A generic-first builder keeps required fields visible, shows source status, and preserves an explicit boundary between AI suggestions waiting in the tray and the doctor’s authored prescription.
Any optional brand sits beside the generic and uses neutral ordering.
An AI candidate for a medicine, test, or dose becomes a prescription item only when the doctor accepts or edits it—and independent checks re-run.
A signature appears only after review, re-authentication, and explicit issue.
1 candidate waiting for doctor review · Source pending · becomes a draft item only when the doctor accepts or edits it.
Operational modules
Encounter capture, draft prescriptions, deterministic checks, referrals, templates, and PDF direction remain usable when connectivity is unavailable. The AI suggestion panel simply shows “unavailable”; the doctor keeps writing. Sync resumes later with conflicts made visible.
Reference views and suggestion cards expose document, edition, page, and corpus version. When the approved corpus does not support a question, the product shows a coverage gap instead of improvising.
Every enabled rule, suggestion source, and dose formula needs a source page, version, clinician author, reviewer, effective date, and expiry. Independent checks are deterministic and run separately from the AI that made the suggestion. This prototype shows the container—never the clinical rule.
Readiness gates
This distinction is the point of the showcase: mechanics can be explored with synthetic data while accountable decisions, licensed sources, and clinical content remain blocked.
Information architecture, empty data models, the suggestion-panel container with empty bundles, offline and sync shells, bilingual rendering, audit mechanics, and research prototypes.
Synthetic data onlyDose ranges and candidate-dose sources, interaction pairs, red-flag logic, suggestion rules, advice libraries, clinical templates, and real formulary content.
Needs CSO, editors, sources, licences, AI evaluation, and DEC-008Real patient storage, pilot use, public claims, and production release remain outside this prototype.
Needs entity, counsel, DPIA, hosting, security, and signed go/no-goStakeholder preview
This static artifact exists to align founders, doctors, safety leadership, design, and engineering on the product shape—including exactly where an AI suggestion stops and the doctor’s decision begins. It is not a clinical tool and is not production-ready.
Return to the interactive preview