Instructions
1. OCR First
Run ~/.venvs/pdf-tools/bin/python3 ~/Code/personal-finance-pipeline/extract_biodata.py . --dry-run. Only use Claude's visual PDF reading for PDFs flagged as failed or low-confidence (especially box-style DOB, height, weight fields).
2. Extract from each PDF
- Applicant Code (top right: AS####, KT####, BMI###, JJ####, DW####, TMV######, EP###)
- Name, DOB (DD-MM-YYYY), Age (calculate from DOB or use stated)
- Height (cm), Weight (kg)
- Children (count and ages if available)
- Religion, Education, Marital Status
- Work Experience (none / local only / overseas), Experience Countries
- Place of Birth, Nationality
Flag missing critical fields (Code, Name, Age, Height, Weight). Cross-check code format against filename. Verify DOB/Age consistency if both present.
3. Update biodata_summary.md
- If file doesn't exist, ask user whether to create one or specify path.
- Match by Code or Name. Only fill empty cells — NEVER overwrite existing data.
- Conflicts with existing data: flag, don't change.
- Backup to
biodata_summary.md.backup-{timestamp} before writing.
- Show proposed changes, confirm with user before applying.
4. Report
Summarize: processed/success/failed counts, per-candidate status, missing data by field, field completeness percentages.
1---2name: extract-biodata3description: Extract applicant biodata from PDF files and auto-populate biodata_summary.md tables. Use when user says 'extract biodata', 'process biodata PDFs', 'update biodata table', or 'scan applicant PDFs'.4---56# Instructions78## 1. OCR First9Run `~/.venvs/pdf-tools/bin/python3 ~/Code/personal-finance-pipeline/extract_biodata.py . --dry-run`. Only use Claude's visual PDF reading for PDFs flagged as failed or low-confidence (especially box-style DOB, height, weight fields).1011## 2. Extract from each PDF12- **Applicant Code** (top right: AS####, KT####, BMI###, JJ####, DW####, TMV######, EP###)13- **Name**, **DOB** (DD-MM-YYYY), **Age** (calculate from DOB or use stated)14- **Height** (cm), **Weight** (kg)15- **Children** (count and ages if available)16- **Religion**, **Education**, **Marital Status**17- **Work Experience** (none / local only / overseas), **Experience Countries**18- **Place of Birth**, **Nationality**1920Flag missing critical fields (Code, Name, Age, Height, Weight). Cross-check code format against filename. Verify DOB/Age consistency if both present.2122## 3. Update biodata_summary.md23- If file doesn't exist, ask user whether to create one or specify path.24- Match by Code or Name. **Only fill empty cells — NEVER overwrite existing data.**25- Conflicts with existing data: flag, don't change.26- Backup to `biodata_summary.md.backup-{timestamp}` before writing.27- Show proposed changes, confirm with user before applying.2829## 4. Report30Summarize: processed/success/failed counts, per-candidate status, missing data by field, field completeness percentages.
hosaypeng