The problem
Doctors lose hours each week typing up visits. Notes get written late in the day from memory, which costs time and accuracy.
An AI scribe has to be right for clinical work: notes need a consistent structure, doctors must stay in control of the final text, and patient information has to be handled with care at every step.
What I built
Record, transcribe, write
The doctor records the visit in the browser. Speech-to-text tuned for medical vocabulary transcribes it, and an LLM writes a structured note that streams onto the screen.
The doctor stays in charge
Every note is a draft until the doctor reviews, edits and finalises it. The AI’s instructions are editable by an admin, so note style can be tuned without a code change.
Privacy safeguards from day one
Two-factor login, automatic logout when idle, account lockout, an audit log of who did what, patient details kept out of system logs, and each doctor’s data kept separate.
Billing that matches usage
Plans and credit packs run on Stripe, so a practice pays for what it uses.
How it runs
The core loop, from the moment something happens to the moment it is handled.
How it runs: Visit recorded, then Transcription, then Clinical note.
What changed
- A recorded visit becomes a structured, editable note without the doctor typing it up.
- Scribeside is now in launch.
He is incredibly proactive, catches edge cases most developers miss, and genuinely cares about the success of the business he is building for.
Under the hood
- Next.js front end
- Node.js (Fastify) API with PostgreSQL and Prisma
- Deepgram medical speech-to-text, with Whisper as an alternative
- LLM note generation through OpenRouter, with streaming
- Stripe plans and credit packs
- Two-factor authentication (authenticator app or email)
- Audit logging and log redaction
Want something like this?
AI development
Summaries, drafting, voice-to-notes and document extraction, added to the product you already run.
SaaS products
AI-native products built to take paying customers from the first release.
Related guides
Guide5 min read
How to hire a full-stack developer for your startup, from the hired side
How to hire a full-stack developer for a startup: what the role covers, how to check real work, the questions that expose gaps, and how to structure pay.
Guide5 min read
How to build an AI agent that survives production
A practical, first-hand guide to building an AI agent: define one job, design the tools, limit its power, test on real examples and control cost.
Questions about Scribeside
Straight answers. If yours is not here, ask me.
Ask me directlyI read and answer every message.Does the AI write the final note?
No. It writes a structured draft. The doctor reviews, edits and finalises every note, and nothing is final until they do.
How is patient information protected?
The system was built with safeguards such as two-factor login, automatic idle logout, audit logs, patient fields redacted from system logs and per-doctor data separation. It has not been certified or audited against any compliance standard.