Fraud Detection Platform - Extraction Accuracy & Expansion

Fraud Detection Platform - Extraction Accuracy & Expansion

Project Title
**Senior Python Developer Needed for Document Fraud Detection Platform (Ongoing)**

---

Project Description

I have an 80% complete document fraud detection platform (Fraud X) built with:
- **Backend**: Python, FastAPI, PostgreSQL, asyncpg
- **Frontend**: React
- **Infrastructure**: DigitalOcean VPS, Nginx, Gunicorn/Uvicorn, HTTPS
- **OCR**: Multi-provider (Google Document AI, AWS Textract, GPT Vision fallback)

Current Status
The core system is working:
- File upload & scan lifecycle
- Multi-provider OCR with scoring
- Fraud engine with PASS/CAUTION/FAIL verdicts
- Admin dashboard with evidence viewer
- JWT authentication & role-based access

What Needs to Be Fixed (Phase 1 - Immediate)

**1. Paystub Extraction Accuracy (~60% of work)**
- Employer name extraction incorrectly captures footer text like "2000 ADP, Inc." instead of actual employer
- Employee name sometimes captures address fields instead of actual name
- Need anchor-based extraction logic for ADP/standard paystub templates
- Need "ignore list" for common footer/header artifacts

**2. Mathematical Validation Issues**
- Current system sometimes shows NetPay > GrossPay (impossible)
- Need pre-validation layer to catch OCR artifacts before they hit the fraud engine

**3. Income Calculation Logic**
- Implement YTD-based monthly income calculation for pay dates after April 1
- Implement current gross × frequency for pay dates before April 1
- Never display $0.00 income - omit field if can't calculate

**4. Questionnaire Integration**
- Wire 7-tier questionnaire results to frontend (currently shows "not available")
- Questionnaire logic exists but isn't being passed to UI

### Phase 2 - Document Type Expansion (Ongoing Work)

Once paystub accuracy is stable, expand to:
1. **Proof of Residence** - Utility bills, lease agreements, bank statements with address
2. **VA/SSI Award Letters** - Government benefit verification
3. **Bank Statements** - Balance verification, income deposits
4. **ID/Driver's License** - Identity verification, expiration checks

Each document type needs:
- Template detection
- Field extraction rules
- Fraud detection rules specific to that document type
- UI integration

---

## Technical Requirements

### Must Have
- Strong Python experience (5+ years)
- Experience with OCR post-processing and text extraction
- Regex and NLP for unstructured text parsing
- FastAPI or similar async Python frameworks
- PostgreSQL
- Understanding of financial documents (paystubs, bank statements)

### Nice to Have
- Experience with Google Document AI or AWS Textract
- Fraud detection or fintech background
- React experience for minor frontend adjustments

---

## Deliverables

### Phase 1 (Fixed Price)
1. Fixed `paystub_builder.py` with accurate employer/employee extraction
2. Mathematical validation layer (pre-check before fraud engine)
3. Income calculation logic per specification
4. Questionnaire data wired to frontend
5. Test coverage for 5+ paystub templates (ADP, UKG, Intuit, Paylocity, Paychex, etc.)

### Phase 2 (Hourly/Retainer)
1. Document type modules for: Proof of Residence, VA/SSI Letters, Bank Statements, ID/DL
2. Ongoing accuracy improvements based on production data
3. ML model training data pipeline setup

---

## Budget

### Phase 1 (Extraction Fixes)
- **Budget**: Fixed price (Make an offer- Negotiable)
- **Timeline**: 1-2 weeks

### Phase 2
- **Budget**: Fixed price (Make an offer- Negotiable)
- **Duration**: 3-6 months initially, potentially longer

---

## How to Apply

Please include in your proposal:
1. Relevant experience with OCR/document processing
2. Example of similar extraction/parsing work you've done
3. Your approach to fixing the employer extraction issue described above
4. Availability and estimated hours for Phase 1
5. Hourly rate for Phase 2 ongoing work

**Bonus**: If you can identify the likely cause of the "2000 ADP, Inc." employer extraction bug just from the description, mention it in your proposal.

---

## Attachments to Include When Posting

- Screenshot of current extraction showing the bug
- Sample Raw OCR output
- Current project structure overview
- Access to staging environment (for serious candidates)

---

## Tags/Skills to Select on Freelancer

- Python
- FastAPI
- OCR
- Document Processing
- PostgreSQL
- Regex
- NLP
- Fraud Detection
- React (secondary)
- AWS Textract
- Google Cloud

---

## Project Type Settings

- **Project Type**: Ongoing project
- **Experience Level**: Expert
- **Estimated Duration**: 1-2 weeks (Phase 1), 3-6 months (Phase 2)
- **Location Preference**: Any (remote)