No description
Find a file
Daniel 9d307fd442
Some checks failed
Forgejo Android APK / Root app tests (push) Successful in 50s
Forgejo Docker Build / Root app tests (push) Successful in 1m2s
Forgejo Android APK / Build signed APK (push) Successful in 2m6s
Forgejo Docker Build / Build Docker image (push) Successful in 37s
Forgejo Docker Build / Deploy to the host (push) Failing after 0s
feat: a vision model looks at the rendered deck and fixes the layout
The model that writes a deck never sees it. It cannot tell that slide four
overflowed, that a nine-item list would read better in two columns, or that two
labelled groups want to be a comparison — those are facts about the rendered
page, not about the text. So each generated deck is now rendered to PDF through
Gotenberg, rasterised to one image per slide with pdftoppm, and shown to a
vision model.

Off unless an administrator names a reviewer, in its own admin card because it
is the one setting that spends money on every generation without a user having
asked for anything. One pass, on generation only: a second pass costs as much as
the first and fixes far less, and refining is a text edit.

It returns a patch, not a deck. Asking for the corrected deck back put the reply
in proportion to the deck rather than to the number of problems, and a
fourteen-slide deck came back cut off mid-object at every output budget the
provider would honour — measured twice before changing shape.

The patch is better for a second reason. The reviewer names a slide and an
action — two columns, one column, split after bullet N, compare with these two
labels — and the server moves the text it already has. The words never pass
through the model, so a review cannot reword, drop or invent a single bullet.
That is a stronger guarantee than instructing it not to and checking afterwards.

The check runs anyway, because a bug in applyChanges would be as bad as a model
rewriting the words and worse for being trusted: body text must come out the
same multiset, figures the same set, and a heading may only be reused or
extended. A continuation heading is the reviewer's one piece of text and is
replaced when it does not continue anything.

Nothing here can fail a generation — no reviewer, an unreachable one, an
unparseable reply, a deck too long to look at, or a patch that applies to
nothing each return the deck that was written.

Verified end to end against a deck with a deliberately overloaded slide: three
slides rendered and sent, one change returned, ten bullets split into five and
five under "Stepwise Management … (continued)", text intact. Left switched off;
enable it under Admin.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU
2026-09-11 21:38:18 +02:00
.forgejo/workflows feat: a deploy you can repeat, and prove afterwards 2026-09-11 00:41:11 +02:00
.github feat: ship reviewed prompt history, conversation limits and account protections 2026-09-07 04:01:01 +02:00
android Fix APK crash: replace XML splash with PNG, add real mipmap launcher icons 2026-03-29 11:07:10 +00:00
assets/learning feat: slides are built by pandoc from markdown, with a reference template 2026-09-11 13:25:24 +02:00
docs feat: a vision model looks at the rendered deck and fixes the layout 2026-09-11 21:38:18 +02:00
e2e fix: voice mode reads the answer that just arrived, and reads what the page shows 2026-09-11 21:10:50 +02:00
migrations feat: sign in with a code emailed to you, offered beside the password 2026-09-11 20:12:03 +02:00
mobile Release v7.14.16 2026-07-31 01:12:47 +02:00
monitoring Add Loki + Grafana monitoring stack, ntfy notifications, biometric auth 2026-04-11 03:00:52 +02:00
public feat: a vision model looks at the rendered deck and fixes the layout 2026-09-11 21:38:18 +02:00
scripts feat: the model designs the deck instead of writing markdown for a parser to guess at 2026-09-11 19:49:55 +02:00
src feat: a vision model looks at the rendered deck and fixes the layout 2026-09-11 21:38:18 +02:00
test feat: a vision model looks at the rendered deck and fixes the layout 2026-09-11 21:38:18 +02:00
.dockerignore feat: ship reviewed prompt history, conversation limits and account protections 2026-09-07 04:01:01 +02:00
.env.example feat: sign in with a code emailed to you, offered beside the password 2026-09-11 20:12:03 +02:00
.gitignore feat: add extension import preview 2026-05-08 22:26:53 +02:00
.gitmessage docs: add CONTRIBUTING.md + .gitmessage template for conventional commits 2026-04-14 23:51:09 +02:00
.node-pg-migraterc.json Add node-pg-migrate for versioned schema changes + better mobile UA labels 2026-04-14 05:06:19 +02:00
admin-cli.js v3.0.0: Auth, admin panel, security fixes, per-tab model selector 2026-03-21 19:25:51 -04:00
CONTRIBUTING.md feat: ship reviewed prompt history, conversation limits and account protections 2026-09-07 04:01:01 +02:00
docker-compose.e2e.yml test(e2e): seed an admin account, and fix the sign-in that broke the browser suite 2026-09-11 17:24:32 +02:00
docker-compose.local.yml feat: ship reviewed prompt history, conversation limits and account protections 2026-09-07 04:01:01 +02:00
docker-compose.monitoring.yml Update bilirubin to exact AAP 2022 values, add exchange transfusion 2026-04-11 05:50:19 +02:00
docker-compose.yml feat: My Resources — anyone can generate teaching material, privately 2026-09-11 14:50:54 +02:00
docker-entrypoint.sh feat: a deploy you can repeat, and prove afterwards 2026-09-11 00:41:11 +02:00
Dockerfile feat: a vision model looks at the rendered deck and fixes the layout 2026-09-11 21:38:18 +02:00
grafana-dashboard.json feat: neonatal calculator, DOCX/PPTX/ODT/EPUB support, gateway-agnostic URL helper, TTS/STT fixes 2026-04-19 02:17:06 +02:00
package-lock.json fix: slides shrink to fit, and an article is never offered as slides 2026-09-11 15:34:16 +02:00
package.json fix: slides shrink to fit, and an article is never offered as slides 2026-09-11 15:34:16 +02:00
README.md docs: My Resources, sign-in codes, invitations, and what the image carries 2026-09-11 20:49:10 +02:00
server.js feat: sign in with a code emailed to you, offered beside the password 2026-09-11 20:12:03 +02:00
TODO.md config: the clinical assistant answers from 12 excerpts 2026-09-11 16:13:55 +02:00

Ped-AI

Ped-AI is a pediatric clinical documentation, education, and bedside decision-support app. This fork has moved well beyond the original scribe app: it now combines encounter documentation, clinical workflows, Learning Hub CMS, admin controls, MCP-backed clinical assistant integration, Redis-backed operational state, and hardened deployment defaults.

The app runs as an authenticated Express/Postgres service with a browser frontend and optional integrations for LiteLLM, Vertex/Gemini, AWS, OpenAI-compatible APIs, Nextcloud WebDAV, S3-compatible storage, OpenBao, Redis, OIDC, TOTP, and Cloudflare Turnstile.

Current Scope

Clinical Documentation

  • Live encounter capture with structured pediatric HPI generation.
  • Dictation cleanup for narrative notes.
  • SOAP, sick visit, well visit, hospital course, chart review, precharting, and ED encounter workflows.
  • Parent-facing education handouts generated from clinician notes, with diagnosis, medication, emergency-care guidance, and preferred-language support.
  • Pediatric developmental milestone tooling.
  • Templates, physician memory, and per-tab model overrides.
  • Server-side speech-to-text routing through configured providers.

Bedside Tools

  • Pediatric calculators and emergency dosing helpers.
  • PE guide and clinical reference content.
  • Vaccines, catch-up schedules, growth/vitals, bilirubin, BSA, GCS, equipment, and resuscitation helpers.
  • Mobile-friendly PWA layout for bedside use.
  • Per-user phone extension and pager directory with soft-delete, search, ZIP export, and JSON/ZIP import for handoff between users.

Learning Hub

  • CMS for articles, clinical pearls, quizzes, and presentations.
  • Tiptap article editor, quiz builder, category management, and draft/publish flow.
  • AI-assisted content generation from topic text, uploaded files, or connected Nextcloud WebDAV files.
  • Marp slide editing with preview and PPTX export.
  • Keyword, semantic, and hybrid search using Postgres/pgvector where configured.

My Resources

  • Private teaching material any signed-in user can generate for themselves — nobody else sees it.
  • Presentations are designed as slide decks (comparisons, tables, callouts, figures beside text), not written as markdown for a parser to guess at.
  • Grounded in the indexed clinical library, and optionally PubMed and the web, each admin-enabled.
  • Optional illustrations, several per resource, placed through the deck.
  • Revise in place, and download as PowerPoint, Word or PDF. See docs/my-resources.md.

Clinical Assistant

  • Optional MCP-backed clinical assistant integration.
  • Prompt suggestions backed by Redis operational cache.
  • No clinical answer response caching.
  • Designed to retrieve from indexed clinical material while keeping provider selection explicit.

Admin And Security

  • Local auth, role-based access, TOTP 2FA, OIDC/SSO, email verification, and optional Turnstile.
  • Admin panel for users, settings, prompts, models, logs, and Learning Hub content.
  • Audit, API, access, and client-error logs with redaction hardening.
  • OpenBao secret loading support at container startup.
  • S3-compatible document storage support.

Removed Browser STT

Browser Whisper has been removed from the runtime. The app should not ship browser Whisper workers, browser-local Whisper model downloads, Transformers.js browser STT, or Browser Whisper setup docs.

Speech-to-text is handled server-side through configured providers such as Google/Gemini, AWS Transcribe, LiteLLM, or OpenAI Whisper. Browser-native Web Speech remains gated behind an explicit user setting when present in the browser.

Quick Start

cp .env.example .env
./scripts/build-image.sh
docker compose up -d --no-build

The default compose exposes the app on 127.0.0.1:3552 and starts:

  • pediatric-ai-scribe for the Node app.
  • pedscribe-db for Postgres with pgvector.
  • ped-ai-redis for operational Redis state.

Health check:

curl -fsS http://127.0.0.1:3552/api/health

Prometheus metrics are exposed at GET /metrics with the ped_ai_ metric prefix.

The first registered user becomes an admin unless registration has already been configured differently.

Core Environment

Set real values in .env before production use.

APP_URL=https://your-domain.example
JWT_SECRET=<64-char-random-secret>
DB_PASSWORD=<strong-database-password>

AI_PROVIDER=litellm
LITELLM_API_BASE=https://your-litellm.example/v1
LITELLM_API_KEY=<key>

TRANSCRIBE_PROVIDER=litellm
LITELLM_STT_MODEL=whisper-1

REDIS_URL=redis://ped-ai-redis:6379

Supported text AI providers include LiteLLM, OpenRouter, AWS Bedrock, Azure OpenAI, and Google Vertex AI. Supported STT routing includes Google/Gemini, AWS Transcribe, OpenAI Whisper, and LiteLLM. Supported TTS routing includes Google Cloud TTS, LiteLLM/OpenAI-compatible audio, and ElevenLabs where configured.

Admin CLI

docker exec pediatric-ai-scribe node admin-cli.js list-users
docker exec pediatric-ai-scribe node admin-cli.js create-admin admin@example.com password123 "Dr. Admin"
docker exec pediatric-ai-scribe node admin-cli.js make-admin user@example.com
docker exec pediatric-ai-scribe node admin-cli.js reset-password user@example.com newpassword
docker exec pediatric-ai-scribe node admin-cli.js toggle-registration
docker exec pediatric-ai-scribe node admin-cli.js stats

Maintenance

The app checks Postgres collation drift on startup and can reindex text indexes after image or OS-library changes.

docker exec pediatric-ai-scribe npm run maint:check
docker exec pediatric-ai-scribe npm run maint:reindex

Run the reindex command after major Postgres image changes, restoring a dump from another distro, or seeing lookup behavior that suggests collation/index drift.

Testing

Run the Node test suite:

npm test

Run syntax checks for touched files when doing focused backend work:

node --check server.js
node --check src/routes/transcribe.js

Run the Playwright smoke suite against the e2e compose stack:

docker compose -f docker-compose.yml -f docker-compose.e2e.yml up -d pediatric-scribe-e2e
npm run e2e

Deployment Notes

  • Put the app behind HTTPS before clinical use.
  • Use only AI/STT/TTS providers covered by your BAA and data-processing requirements.
  • Configure OIDC/SSO and 2FA for production users.
  • Keep JWT_SECRET, database credentials, provider keys, S3 keys, SMTP credentials, and OpenBao tokens out of git.
  • Treat logs as sensitive operational data even with redaction enabled.
  • Use the Caddy/reverse-proxy layer to expose only intended public routes.

Documentation

Primary references:

  • docs/ARCHITECTURE.md for the current system map and service boundaries.
  • docs/DEVELOPMENT.md for day-to-day code-change workflow.
  • docs/SCALING.md for scaling priorities and readiness work.
  • docs/CLINICAL_ASSISTANT.md for MCP-backed assistant behavior and safety rules.
  • docs/MODULE_CONVENTIONS.md for CommonJS, ESM, globals, and rendering rules.
  • docs/architecture.md for high-level architecture.
  • docs/api-reference.md for API routes.
  • docs/authentication.md for auth, OIDC, and security configuration.
  • docs/ai-providers.md for model/provider setup.
  • docs/speech.md for server-side STT/TTS setup.
  • docs/learning-hub.md for the CMS and education workflow.
  • docs/my-resources.md for private teaching material, the slide renderer, and search sources.
  • docs/retrieval-tuning.md for how much corpus each feature retrieves, and what it costs.
  • docs/configuration.md for environment variables.
  • docs/deployment.md for production deployment.
  • docs/mobile-build.md for the Capacitor wrapper and app-store build notes.
  • docs/logic/README.md for the deeper code walkthrough.

Some deep docs/logic/ files still describe historical implementation details. Prefer runtime code and tests when documentation conflicts with current behavior.

Clinical Safety

Ped-AI is documentation and education support software. It does not replace clinical judgment, local policy, medication verification, or attending review. Validate generated notes, calculations, and recommendations before use in patient care.