AI/MLFull-Stack

Mdar Loan Estimation API

ML-backed loan estimation service for real estate funding

Built an ML inference API for loan estimation using scikit-learn models and FastAPI deployment to support product decisioning workflows.

Overview

Mdar Loan Estimation API is a machine learning service that estimates loan-related values for a real estate funding workflow.

Problem

The funding workflow required consistent, explainable predictions rather than manual estimation.

Solution

I developed preprocessing and regression pipelines using scikit-learn (including Ridge and HistGradientBoostingRegressor), applied log transforms and constraints, and exposed predictions via FastAPI with serialized model artifacts.

Tech Stack

Python
pandas
NumPy
scikit-learn
FastAPI
joblib

Architecture

Simplified flow diagram rendered as text.

Client / Platform
  -> FastAPI /predict endpoint
    -> Input validation + preprocessing pipeline
      -> Trained model (Ridge / HistGradientBoostingRegressor)
        -> Prediction response
  -> joblib model persistence

Key Features

Structured preprocessing and feature preparation pipeline
Multiple regression model strategy for comparison
FastAPI endpoint for production-style prediction requests
Model artifact persistence with joblib

Challenges & Learnings

Improved stability by applying transformations and constraints on prediction outputs.
Learned how to package ML pipelines for API-first product integration.
Reinforced best practices around reproducibility and model serving boundaries.

Screenshots

Links

Private repo