Agri-Access: AI-Powered Agricultural Credit Scoring

APRU Tech Policy Hackathon 2025 - Deliverable #3 README

High-level Explanation

Agri-Access uses satellite imagery and AI to assess agricultural credit risk in about one minute, enabling Indonesian banks to serve 29 million farmers currently excluded from formal credit.

How It Works

1. Satellite Analysis: IBM/NASA Prithvi foundation model processes real-time NASA satellite imagery of farm locations
2. AI Risk Assessment: Machine learning combines satellite data with weather and farm information to predict credit risk
3. Banking Integration: Generates Basel III-compliant risk parameters and Indonesian SLIK credit scores (1-5 scale)
4. Farmer Explanations: Google Gemini AI provides clear recommendations

Key Innovation

• Real-time processing: ~1 minute credit assessment vs weeks traditionally
• No field visits required: Satellite coverage works anywhere in Indonesia
• Regulatory compliant: Basel III + Indonesian banking standards ready
• Transparent AI: SHAP explanations show decision factors

Setup and Use Instructions

Prerequisites

• Python 3.8+
• Internet connection for satellite data

Installation & Demo

# 1. Navigate to project directory
cd agri-access

# 2. Install dependencies
# Public deploy/lightweight:
pip install -r requirements.txt

# Full local private platform:
# pip install -r requirements-private.txt

# 3. Start application
python3 basel_iii_api.py

# 4. Open browser to the displayed localhost website

Demo Instructions

1. Select demo farmer: Click Any Demo Data for pre-configured scenarios
2. Run analysis: Click "Analyze Credit Risk" (~1 minute processing)
3. View results: Credit score, satellite imagery, AI explanations, risk parameters

Optional: Protect API Access

To avoid exposing backend APIs publicly, set a token:

export AGRI_API_AUTH_TOKEN="your-strong-demo-token"

Then open platform with a tokenized link once (sets an auth cookie for browser API calls):

http://localhost:3000/platform?token=your-strong-demo-token

For direct API clients, send either:

• Authorization: Bearer your-strong-demo-token, or
• X-API-Key: your-strong-demo-token

Public Demo vs Private Platform

• Public demo deploy: keep AGRI_ENABLE_PRIVATE_PLATFORM=false (default). Only the feasibility landing demo is exposed.
• Private local platform: run locally with:

export AGRI_ENABLE_PRIVATE_PLATFORM=true
python3 basel_iii_api.py

This enables /platform and all /api/* routes only in your local/private environment.

Demo Scenarios

• Rice farmer (West Java) - Good credit example
• Palm oil farmer (Sumatra) - Excellent credit example
• Coffee farmer (Central Java) - Fair credit example

AI Tool Disclosure

AI Tools Used in Development

• Claude Code (Anthropic): Code development, debugging, implementation assistance
• Google Gemini: Document polishing, research assistance for Indonesian banking regulations

AI Tools Integrated in Product

• IBM/NASA Prithvi-EO-2.0-300M: Satellite imagery analysis (foundation model)
• Google Gemini 2.5 Flash: Real-time farmer explanations in Indonesian/English

Original Work (Not AI-Generated)

• Core innovation concept (satellite-based agricultural credit scoring)
• System architecture and technical implementation approach
• Basel III compliance and risk assessment
• API integrations (NASA, weather services, Gemini)
• Indonesian banking compliance

All AI-generated code was reviewed, tested, and customized for Indonesian agricultural finance requirements.