Trustline — A Wells Fargo ML Foresight System

Origin — The I-75 Moment

Millions of people live one unexpected bill away from crisis. As an international student driving a failing car down I-75 in Michigan, every engine knock felt like a financial cliff. One breakdown could have ended everything.

That moment became the seed for Trustline — a system built not just to offer cash, but to replace anxiety with foresight. Trustline isn’t about quick fixes. It’s about building stability through behavior, coaching, and earned support.

---

Mission Statement

"Trustline transforms a bank from a last-resort lender into a partner in progress."

Instead of reacting to overdrafts and crisis moments, Trustline predicts risk before it happens, nudges users toward better choices, and rewards discipline with Coach Credit — a living trust signal that unlocks fair, fast micro-advances.

This repo contains the first working ML prototype of that system.

---

What This Repo Demonstrates

This GitHub repository hosts a logistic regression model designed to:

• Ingest anonymized transaction data
• Engineer time-sensitive behavioral features (late-night spending spikes, irregular withdrawals, risky transaction categories)
• Predict overspending risk within the next 3 hours
• Trigger context-aware nudges, similar to production behavior

This is a lightweight, explainable ML system — deployable with current Wells Fargo infrastructure and built to scale.

---

System Architecture — High Level

1. Data Ingestion

• Uses CSV/transaction logs (expandable to Wells Fargo’s internal feed)
• Categorizes transactions with nuanced weighting (essentials vs. discretionary)

2. Feature Engineering

• Extracts behavioral signals like:
  • spend_velocity_change
  • late_night_discretionary_ratio
  • goal_alignment_score

3. Risk Prediction

• Logistic Regression (explainable, regulator-friendly)
• Outputs a probability of crisis within a 3-hour prediction window

4. Nudge Generation

• Risk < 30% → Positive reinforcement nudge
• 30%–60% → Gentle redirection
• 60% → Urgent alert + Coach Credit at stake

5. Micro-Advance Gateway

• If Coach Credit ≥ threshold → Instant cash access via existing rails
• Decision backed by both FICO and a real-time behavioral trust score

---

Coach Credit — Trust You Can See

Coach Credit isn’t like FICO. It's dynamic and behavior-based, designed to show who a user is becoming, not just who they were.

Actions like skipping a risky purchase after a nudge or moving $20 to savings increase Coach Credit immediately. This makes progress feel real and motivating.

---

How to Run Locally

# Clone the project
git clone https://github.com/your-handle/trustline-ml
cd trustline-ml

# Optional: create a virtual environment
python3 -m venv .venv
source .venv/bin/activate

# Install dependencies
pip install -r requirements.txt

# Run the simulation demo
python run_demo.py