Record Review Copilot (Demo)

A tiny Rails app that uploads a text-based medical PDF, extracts text, calls an LLM to auto-summarize key facts (diagnoses, medications, procedures, dates, providers), lets an analyst edit the results inline, and exports a CSV.

Stack

• Ruby on Rails, Postgres, Tailwind CSS, Hotwire, Svelte, Heroku (Salesforce-ready later).

⚠️ Demo-only / HIPAA disclaimer This repository is not production-ready. It does not meet HIPAA compliance requirements. Do not upload real PHI. It sends text chunks to a third-party LLM API. Logs and error traces may include data. A production version would require a BAA with the LLM vendor, strict logging/retention controls, encryption at rest/transport, access controls, audit logging, and a security review.

Feature Summary

• Upload PDF → extract text (text-based PDFs only; OCR not included).
• Chunk text → call LLM per chunk with a strict JSON schema.
• Merge/dedupe findings across chunks.
• Inline editable findings table (Svelte + Stimulus/Hotwire), save per row.
• Export CSV of findings.
• Optional redact mode to avoid echoing names/addresses in model output (demo-level).

Models

• Document(id, title, status, text, redact:boolean)
• Finding(id, document_id, category, label, value, date, confidence:float)

Categories returned by the LLM

• diagnoses, medications (name,dose), procedures (text,date), key_dates (label,date), providers (name,type)

Architecture (high-level)

• Controllers: DocumentsController (new/create/show/analyze/export), FindingsController#update
• Services:
  • Pdf::ExtractText — text extraction via pdf-reader
  • Text::Chunker — 3–5k char chunks
  • Llm::MedicalFactsExtractor — OpenAI Chat Completions → strict JSON
  • Findings::Merge — dedupe by normalized keys
  • Csv::Exporter — single CSV across categories
• Frontend:
  • Tailwind (cssbundling)
  • Svelte component mounted via Stimulus for editable table

Live Heroku App

• Navigate to 'https://record-review-copilot-6b33aa679088.herokuapp.com/'
• Fill out form and upload a sample medical document pdf (note that this app is for demo purposes and is looking for specific keywords that may not be optimized for real docs yet)
• Click "Analyze" to fetch and save findings from the LLM results
• Click "Export CSV"

Local Development

Prereqs

• Ruby 3.2.x
• Postgres
• Node ≥18 <23, Yarn ≥1.22
• OpenAI API key

Setup

# install Ruby gems and JS deps
bundle install
yarn install

# DB
bin/rails db:create db:migrate

# .env
cat > .env <<'ENVVARS'
OPENAI_API_KEY=sk-...
OPENAI_MODEL=gpt-4o-mini
ENVVARS

# build assets
yarn build && yarn build:css

# run dev (Rails + Tailwind + esbuild)
bin/dev