What It Does
The pipeline takes a client's raw blood test PDF and processes it automatically. The NP's review time drops from ~45 minutes per client to ~5 minutes.
| # | Step | Model / Tool | Output |
|---|
| 1 | Ingest | — | Raw PDF saved, run created |
| 2 | Extract text | pdf-parse → Claude vision fallback | Clean text |
| 3 | Parse biomarkers | Claude Sonnet 4.6 | Structured JSON |
| 4 | Clinical analysis | Claude Opus 4.7 | Summary + recommendations |
| 5 | Generate client PDF | Puppeteer | Branded PDF report |
| 6 | Write to GHL | GHL REST API | Updated contact |
Cost Per Report
| Step | Model | Est. Cost |
|---|
| OCR (if scanned PDF) | Sonnet 4.6 vision, ~5 pages | ~$0.05 |
| Parse biomarkers | Sonnet 4.6 | ~$0.02 |
| Clinical analysis | Opus 4.7 | ~$0.10 |
| Total | | ~$0.15–0.20 |
GHL Tag Lifecycle
What Gets Written to GHL
| Field / Action | Value |
|---|
contact.call_notes | AI-generated clinical summary (NP edits here) |
| Tag added | bloodwork_completed |
| Tag removed | bloodwork_scheduled |
| Note created | Full structured analysis with flagged markers and recommendations |