Project Story: CodeSure – Chrome Extension for CPT Validation

Inspiration

Claim denials waste time and money—often because codes or payer rules are misunderstood at the moment of entry. In this hackathon I set out to build a lightweight, privacy-first helper that lives in the browser and delivers on-device intelligence for fewer errors, faster approvals, and a calmer day for clinicians and billers.

What it does

CodeSure helps you go from service description → suggested codes → payer rules → prior-auth note in one flow:

• AI Assist (Prompt API): Type “CT abdomen with contrast” and get ranked suggestions from a local CPT-like mock catalog; attachment filenames act as hints.

• Validate & Compare: Select a payer to see rule text, or open a side-by-side payer table. Click Explain rule for a one-sentence rationale (Prompt API). Export CSV or copy as text.

• Policy Summarizer (Summarizer API): Summarize the open policy page into key points; Translator API renders outputs in EN/ES/JA. Falls back to Prompt API if Summarizer isn’t available.

• PA Note Builder: Structured fields (Medical Necessity, History/Imaging, Site/Modifiers) with AI drafts and minimal polishing, then assemble a clean plain-text PA note.

• Coverage Snapshot: One-click digest of payer rules plus top required docs inferred from your summary.

• Diagnostics: Built-in checks for models and data to keep demos reliable—even offline after warm-up.

• Persistent History (popup-only): Recent suggestions/notes are saved for quick recall, with a Clear History control.

• Cross-tab Autofill: Sync from Home brings patient/service details into the PA Note and auto-fills relevant fields.

• Language Pack Everywhere: Home and PA Note UIs are translatable (EN/ES/JA) so the whole flow can localize.

• Quality-of-life: Paste from Capture pulls the last captured text into the prompt; Clear buttons reset the Home and PA Note forms; a progress bar shows Check Auth Status activity.

How I built it

• Manifest V3 Chrome extension with a secure popup UI (strict CSP, no inline scripts). The footprint is popup-only—the previous “Open in tab” window was removed.

• On-device Chrome AI:

• Prompt API (LanguageModel) for attribute extraction, suggestion ranking, and the Explain rule rationale.

• Summarizer API for policy key points with Prompt API fallback and timeouts.

• Translator API for multilingual output (EN/ES/JA) with a Prompt fallback if needed.

• Workers & Scripts: A fast suggestion Web Worker over generated mock data (tools/generate-mock-cpt.js, tools/generate-mock-rules.js). A background service worker coordinates long tasks and storage; a content capture script enables Paste-from-Capture.

• State & Storage: Persistent session history, cross-tab sync for autofill, and explicit Clear/Clear History controls.

• UX & A11y: Keyboard-first navigation, visible focus, progress indicators, CSV export, and helpful empty/error states.

• Resilience: A Diagnostics panel validates Data/Worker/LanguageModel/Translator/Summarizer before demoing; graceful degradation when an API isn’t available.

Challenges I ran into

• Model availability & cold-start: Managing downloads, timeouts, and fallbacks without user confusion.

• CSP & event handling: Eliminating inline handlers while keeping the UI snappy.

• Suggestion recall: Ensuring breadth across imaging, E/M, labs, DME, vision, anesthesia, therapy, pathology with domain synonyms.

• Localization: Keeping labels, messages, and outputs consistent across screens (Home + PA Note).

• History design: Storing enough context to be useful while keeping cleanup obvious and one-click.

Accomplishments that I'm proud of

• A complete on-device, privacy-first workflow from description to payer-ready note—inside the browser.

• Explain rule turns opaque policy text into a clear rationale, inline.

• Coverage Snapshot + CSV export that feel practical for real teams.

• Reliable demo: Offline after warm-up, diagnostics green, and graceful fallbacks.

• A cleaner architecture: popup-only UI, background service worker, capture script, persistent history.

What I learned

• How to structure and ship a Chrome extension (MV3) with a small, safe surface area.

• Practical regex validation patterns for healthcare-style codes.

• Balancing usability, speed, and accuracy in real-time workflows.

• Designing non-intrusive alerts and progress feedback that build trust.

• Why explicit Clear/Clear History affordances matter when saving user context.

What's next for CodeSure

• Swap mock data for licensed CPT + payer policy sources via user keys.

• Add ICD-10/HCPCS, payer-specific templates, and more languages.

• Multimodal intake (OCR/PDF parsing) with on-device pre-processing.

• Optional hybrid AI (Firebase AI Logic / Gemini API) for heavy tasks while preserving privacy defaults.

• Export to standards (e.g., FHIR) and add an admin dashboard with analytics and compliance guardrails.

Built With

• css
• html5
• javascript
• json
• node.js
• promptapi
• regex
• summarizerapi
• translatorapi