Banyan Bat: Tanaji: Teaching an AI to Fight Fraud Like a Warrior

🔥 Introduction

If there's one name in Maratha history that embodies courage under pressure, it’s Tanaji Malusare.

When the Kondhana Fort had to be reclaimed, Tanaji didn’t wait. He scaled the cliffs with a monitor lizard 🦎, led a night raid, and paid the ultimate price — but won the fort.

In the Fraud-Free Swarajya project, Tanaji is our AI fraud scorer — making fast, high-stakes decisions on every incoming transaction.

Just like the real Tanaji, our AI:

Works under uncertainty
Evaluates risk on the fly
Doesn’t wait for permission to act 🚀

🧠 What Tanaji (the AI) Does

The tanaji module is a Python-based microservice powered by:

✅ FastAPI
✅ scikit-learn for ML
✅ pydantic for validation
✅ uvicorn for async web serving
✅ Logging + formatting with loguru and pythonjsonlogger

It receives a transaction like this:

json

{
  "transactionId": "txn001",
  "userId": "user123",
  "amount": 5000,
  "paymentMethod": "risky",
  "location": "risky"
}

Then it returns a fraud score, a risk level, and even an explanation.

🧪 How It Works Behind the Scenes

Feature Extraction:
We extract meaningful inputs like:
- Amount
- Whether the location or payment method is risky
Model Training:
We trained a basic RandomForestClassifier on synthetic data to:
- Classify transactions as fraud or not
- Assign a confidence score (0 to 1)
Scoring Endpoint:
Exposed at /score, it predicts and responds in milliseconds.
Logging Everything
Every request and response is structured, logged, and dockerized.

🧱 Why FastAPI?

Because:

It’s async by design
Lightning fast (even for ML)
Auto-generates Swagger docs
Works beautifully inside Docker
And let’s be honest: It feels like Spring Boot’s cousin — but cooler with Python 🐍

🔥 Tanaji Trivia Time

Here are some legendary facts about Tanaji Malusare:

💥 He led the assault on Kondhana Fort in 1670 with just a handful of soldiers.
🦎 He used a monitor lizard (ghorpad) to scale the vertical walls of the fort at night.
💔 Though mortally wounded in the battle, his mission succeeded — and Shivaji said:
"Gad aala, pan Sinh gela." (We got the fort, but lost the lion.)
🗡️ Kondhana was later renamed Sinhagad in his honor.

Like Tanaji, our AI doesn't hesitate. It detects danger, acts swiftly, and protects the digital kingdom.

⚙️ Sample Response from Tanaji

json

{
  "transaction_id": "txn001",
  "fraud_score": 0.97,
  "explanation": ["High risk score", "High amount"],
  "risk_indicators": ["HIGH_AMOUNT"]
}

Simple. Actionable. Defensible.

🧰 Dockerizing Tanaji

We packaged Tanaji with a clean Docker setup:

uvicorn + FastAPI as entrypoint
Exposes port 8000
Shared volume for fraud_model.pkl
Ready to scale with Docker Compose

dockerfile


CMD ["uvicorn", "app.main:app", "--host", "0.0.0.0", "--port", "8000"]

🧠 What Makes This Exciting?

You can retrain the model (train_model.py) anytime
Logs are structured and Prometheus-ready
Pluggable with any orchestrator or microservices platform
As the Fraud-Free Swarajya grows, Tanaji will evolve — just like his legacy lives on