A complete, production-shaped clinical dictation-to-sign-off system, Next.js, NestJS, PostgreSQL, and AI transcription, delivered as real, unlocked source code you can run locally, study, and build on.
The current state: how clinical documentation actually gets built today
Every hospital, clinic, and multi-provider practice runs on documentation. A doctor sees a patient, dictates or writes notes, and somehow that spoken or scribbled information has to become a clean, structured, permanent medical record, one that a transcriptionist can work from, a quality reviewer can check, a physician can sign off on, and an auditor can trust years later. That sounds simple in a sentence. In practice, it is one of the more complex software problems in healthcare technology.
For teams that want to build or evaluate a system like this, the usual starting points are not great. One path is buying an established, closed clinical documentation platform, which usually means a long procurement cycle, a recurring per-seat subscription, and a black box you cannot inspect, modify, or self-host. Another path is building from scratch, which means months of architecture decisions before a single note gets transcribed: how do you model roles and permissions for doctors, transcriptionists, and QA reviewers? How do you version a note so every edit is auditable? How do you wire up AI transcription without locking yourself into one vendor forever? A third, increasingly common path is stitching together a generic transcription API with a homemade admin panel, which tends to fall apart the moment real clinical workflow requirements show up, routing, review stages, audit logs, role-based access.
Developers, health-tech founders, agencies, and technical teams inside hospitals or clinics who want to prototype, demo, or actually deploy a documentation workflow are left choosing between an expensive closed product and a long, risky build from a blank repository. There has not been an obvious middle path: real, working, inspectable source code for the whole workflow, ready to run today.
The pain point: the parts that are hard to get right are the parts everyone has to build from zero
The friction is not in any single screen. It is in the connective tissue that a real clinical documentation workflow needs, the pieces that are tedious and easy to get wrong, and that every team ends up reinventing separately:
- Role-based access control that actually reflects a hospital. A generic "admin vs. user" permission model does not map to how a hospital works. Doctors, transcriptionists, QA reviewers, department managers, staff, and super admins each need different views, different actions, and different data visibility, and getting that wrong is both a workflow problem and a compliance risk.
- A multi-stage review pipeline. A clinical note is not "done" the moment AI drafts it. It needs to flow from draft, to transcriptionist cleanup, to QA review, to doctor approval, with a clear status at every stage. Building that state machine correctly, and making it visible to every role, takes real engineering time.
- Auditability and versioning. Every edit to a medical record ideally needs a trail: who changed what, and when. Retrofitting version history and audit logging onto a system that was not designed for it from day one is painful and error-prone.
- AI transcription without vendor lock-in. Wiring a cloud transcription API into a prototype is easy. Building it so the engine is swappable, so a team can run it fully offline with a self-hosted model instead of sending patient audio to a third party, is a much bigger architectural decision, and one most teams do not make until it is too late.
- Everything else a real product needs. Notifications, search, per-department dashboards, OCR for scanned lab reports, a queueing system for background jobs, none of these are the "interesting" part of a health-tech build, but skipping them is exactly what turns a demo into something that cannot be shown to a real hospital operations team.
The net effect is that a huge amount of engineering time on clinical documentation projects goes into scaffolding, before a team ever gets to the part that is specific to their product. That is expensive in developer hours, and it is expensive in time to a working demo, which matters enormously when you are trying to validate an idea, pitch a client, or evaluate a build-vs-buy decision.
The solution: MediScribe AI, the full platform as real source code
MediScribe AI from HolyByte Innovations is complete, production-shaped source code for a hospital medical transcription platform, the full stack, ready to run locally in minutes. This is not a locked SaaS product and not a stripped-down demo. It is a Next.js 14 (App Router) and TypeScript frontend, a NestJS and TypeScript backend, and PostgreSQL via Prisma for the data layer, delivered with no obfuscation and nothing held back.
Critically, it ships MIT licensed, with no license key and no phone-home activation. That means you can use it, modify it, deploy it, and build your own service on top of it, on your own terms. It is designed to run on your own machine with a simple command set, and it includes a zero-install local Postgres setup so you do not need Docker just to get it running and see the workflow end to end.
At its core, MediScribe AI models an end-to-end clinical dictation-to-sign-off workflow: a doctor dictates or uploads audio, AI drafts the note, the note auto-routes to a transcriptionist, then moves to QA review, then to doctor approval, and ends as a versioned, audit-logged record with one-click PDF export. That is the exact pipeline described above under "the pain point," already built, already wired together, and already testable with seeded demo accounts across every role.
A clear, honest note up front: MediScribe AI is developer source code intended for evaluation and customization, not a certified medical device. A real clinical deployment needs the appropriate compliance and security review for your region before it touches real patient data. This post is describing the software as shipped; it is not medical advice or a claim of regulatory certification.
Key features: what is inside the platform
Seven roles with real role-based access control
MediScribe AI ships with seven distinct roles, admin, doctor, transcriptionist, QA reviewer, department manager, staff, and super admin, each with its own permissions and views. Seeded demo accounts are included for every role, so you can log in as a doctor, then as a transcriptionist, then as a QA reviewer, and see exactly how the same note looks and behaves from each seat. That is the fastest way to evaluate whether a role model fits your own organization's structure, without writing a single line of code first.
AI transcription with a swappable engine
The platform includes AI transcription built around a swappable, on-premise Whisper engine. You bring your own backend, whether that is a free self-hosted model or a paid API, so you are never locked into one vendor's pricing or data-handling terms. It ships with a safe offline transcription stub by default, so you can explore the full workflow immediately, and a docker-compose service and setup guide are included for pointing it at a real self-hosted Whisper (large-v3) server when you are ready for production-grade accuracy.
Confidence-based triage and offline medical NLP
Not every AI transcription is equally reliable, and MediScribe AI does not pretend otherwise. It includes confidence-based "needs review" triage, so lower-confidence transcriptions are flagged for closer human attention rather than silently passed through. On top of that sits an offline medical NLP pass that performs drug detection and suggests ICD-10 and CPT codes, giving transcriptionists and reviewers a useful starting point instead of a blank note.
On-device OCR for scanned documents
Clinical workflows are not audio-only. MediScribe AI includes on-device OCR, so a scanned lab report can be pulled straight into the note rather than retyped by hand, another place where the platform handles a realistic, messy piece of clinical documentation instead of assuming a clean, ideal input.
Real-time notifications and an AI learning loop
The system includes real-time notifications so staff know when a note needs their attention as it moves through the pipeline, plus an AI learning loop that improves from human edits over time, using the corrections transcriptionists and reviewers make as a feedback signal rather than throwing that information away.
Dashboards, global search, and versioned, audit-logged records
Operationally, MediScribe AI includes per-department workload and operations dashboards and global search across notes, so managers and staff can find and monitor documentation at scale rather than hunting through individual records. Every note that completes the workflow becomes a versioned, audit-logged record with one-click PDF export, the kind of traceability and portability a real clinical record needs.
A modern, well-tested technical stack
Under the hood, MediScribe AI is built on Next.js, NestJS, TypeScript, PostgreSQL and Prisma, Tailwind, and Socket.IO, with an optional Redis and BullMQ queue for background jobs and S3-compatible storage (local disk by default, so nothing external is required to get started). It ships with 68 automated tests included, and both the frontend and backend apps type-check and build clean, which matters enormously when you are evaluating source code you plan to extend rather than just read.
Who it's for
- Health-tech developers and founders who want a real, working starting point for a clinical documentation product instead of months of scaffolding before the first feature.
- Software agencies evaluating or pitching a transcription or documentation workflow to a healthcare client, who need something concrete and demoable rather than a slide deck.
- Hospital and clinic IT or innovation teams doing a build-vs-buy evaluation, who want to see a full role-based workflow running locally before committing budget either way.
- Engineers who want to learn from a realistic, full-stack TypeScript codebase, with a genuine multi-role permission system, a queue-backed AI pipeline, and audit logging, all patterns that show up far beyond healthcare specifically.
- Teams that care about data control and want the option to run transcription fully offline, on their own infrastructure, rather than sending audio to a third-party cloud service by default.
How it works, step by step
The workflow modeled inside MediScribe AI follows the same path a note takes in a real documentation department:
- Capture. A doctor dictates a note or uploads existing audio through the frontend.
- AI draft. The configured transcription engine, whether the default offline stub or a connected self-hosted Whisper server, produces a first-pass draft of the note.
- Confidence triage. Lower-confidence sections are flagged for review rather than silently accepted, and the offline medical NLP pass suggests relevant drug mentions and ICD-10/CPT codes.
- Transcriptionist routing. The draft auto-routes to a transcriptionist for cleanup and correction.
- QA review. A QA reviewer checks the cleaned note against quality expectations before it moves forward.
- Doctor approval. The originating doctor reviews and signs off, finalizing the record.
- Versioned, audit-logged output. The finished note is stored as a versioned record with a full audit trail and can be exported to PDF in one click.
Because every stage is backed by real role-based accounts, you can walk this entire path yourself using the seeded demo logins, seeing exactly what each participant sees, before you write any custom code.
Real ways it saves time
The most direct benefit is skipping the scaffolding phase entirely. Instead of spending weeks designing a role model, a review pipeline, and an audit-logging scheme from a blank repository, you start from a version that already works, already has demo data, and already type-checks and builds clean with automated tests in place. That turns "let's evaluate whether this architecture fits us" from a multi-week research project into an afternoon of running the app locally and clicking through it as each role.
The second benefit is architectural flexibility without vendor risk. Because the AI transcription engine is swappable and the license has no activation or phone-home requirement, you are not building your roadmap around a vendor's pricing changes or API deprecations. You can start with the included offline stub, move to a self-hosted Whisper server using the provided docker-compose setup, or integrate a different provider later, all without re-architecting the rest of the system.
The third benefit is that the unglamorous but essential pieces, OCR for scanned reports, real-time notifications, per-department dashboards, global search, background job queueing, are already present. Those are exactly the features that are easy to underestimate in a project plan and expensive to retrofit later, and here they come included in the base platform.
Tips to get the most out of it
- Start with the seeded demo accounts. Log in as each of the seven roles before writing any custom code, so you understand the full workflow from every seat first.
- Keep the offline stub while you explore. There is no need to stand up a real Whisper server on day one; the safe default lets you evaluate the entire pipeline first.
- Read the README before your first run. The full setup steps, including the docker-compose service for a production-grade transcription engine, are documented there.
- Run the included test suite early. With 68 automated tests already in place, running them right after setup is a fast way to confirm your local environment matches what is expected before you start customizing.
- Plan your compliance review separately from your technical evaluation. Treat MediScribe AI as the technical foundation, and scope the regulatory and security review your region requires before any real clinical deployment.
Frequently asked questions
What exactly do I get with MediScribe AI?
You get the complete source code for a hospital medical transcription platform: a Next.js 14 and TypeScript frontend, a NestJS and TypeScript backend, and a PostgreSQL database via Prisma, delivered as a single .zip with a full setup guide in the README. Nothing is obfuscated and nothing is held back.
Is this a SaaS subscription or a license key product?
No. MediScribe AI is MIT licensed with no license key and no activation or phone-home requirement. You use, modify, and deploy it on your own terms, including building your own service on top of it.
Do I need Docker to try it?
No Docker is required just to get started, since it includes a zero-install local Postgres setup. Docker is used optionally later, through an included docker-compose service, if you choose to connect a self-hosted Whisper server for production-grade transcription accuracy.
Is MediScribe AI a certified medical device?
No, and this is stated plainly by design. It is developer source code intended for evaluation and customization. Any real clinical deployment needs the appropriate compliance and security review for your region before it is used with real patient data.
Can the AI transcription run fully offline?
Yes. The transcription engine is swappable and built around an on-premise Whisper model. It ships with a safe offline transcription stub by default, and a docker-compose service and setup guide are included for connecting a self-hosted Whisper (large-v3) server for production accuracy, so patient audio does not need to leave your own infrastructure.
Who is this built for?
Developers, health-tech founders, agencies, and technical teams inside hospitals or clinics who want a real, working foundation for a clinical documentation workflow, whether that is for a prototype, a client demo, or a serious build-vs-buy evaluation.
Watch more from HolyByte on YouTube
Product demos, tutorials and walkthroughs from our channel
Ready to try MediScribe AI?
Get the complete, unlocked source code for a hospital medical transcription platform and see the full role-based, AI-assisted workflow running on your own machine.
Get MediScribe AI on the HolyByte Store
Join the HolyByte Affiliate Program
Love HolyByte products? Earn a commission by referring them to your audience, friends, or community.
Become a HolyByte Affiliate
Follow & subscribe
Stay close to HolyByte Innovations for product news, guides, and launches.
LinkedIn ·
Facebook ·
YouTube
Comments
Post a Comment