Open to opportunities

Udit Pahwa

@uditpahwa

Message

I’m a Data Scientist building agentic AI and credit models that lift conversion, risk accuracy, and deployment speed.

India

Message

What I'm looking for

I’m looking to build and deploy impactful ML/GenAI systems in credit and risk contexts—combining agentic pipelines, rigorous experimentation, and strong engineering discipline with mentorship and fast iteration.

I’m a Data Scientist who turns unstructured data and credit-domain signals into production-ready machine learning and GenAI systems. At Khatabook, I architected an agentic multimodal framework using category-specific prompts and RAG, deployed via FastAPI, expanding feature coverage from 60% to 90% at 99%+ accuracy.

I also build end-to-end underwriting and decisioning engines: an ensemble credit underwriting engine with 8 models and K-Means improved conversion by 15%, and an offer model deployed on Kubernetes increased overall portfolio yield by 12%. Previously at Fi Money and ANZ Bank, I delivered propensity and risk scoring improvements (3x conversion; +15% predictive power; +10% relative Gini), reduced batch time by 50%, and lowered false positives by 25%, while mentoring and accelerating model deployment cycles.

Experience

Work history, roles, and key accomplishments

Current

Data Scientist II

Current

Khatabook

Jul 2024 - Present (2 years)

Architected an agentic multimodal framework to parse unstructured B2C data into structured signals using category-specific prompts and RAG, deployed via FastAPI (feature coverage 60% to 90% at 99%+ accuracy). Built a credit underwriting engine using an 8-model ensemble with K-Means and an offer model for optimizing loan terms; led a team of 10 and improved portfolio yield by 12%.

Agentic AI Prompt Engineering RAG fastAPI Credit Underwriting Kubernetes Feature Engineering Python R Machine Learning A B Testing

Senior Data Scientist

Fi Money

May 2023 - Jul 2024 (1 year 2 months)

Developed a loan propensity model using an XGBoost classifier validated with hypothesis testing and A/B testing, deployed as a microservice to drive 3x higher conversion in top segments. Built a scalable feature computation pipeline using PySpark and SQL orchestrated via Airflow (50% reduction in daily batch time) and created an anomaly detection system for transaction monitoring (25% fewer false

XGBoost Microservices A B Testing Hypothesis Testing Feature Engineering Pyspark SQL Airflow Linear Algebra Statistics Neural Networks BigQuery Anomaly Detection

Data Scientist

ANZ Bank

Jun 2021 - May 2023 (1 year 11 months)

Rebuilt the acquisition risk scorecard by engineering 1,000+ features from structured bureau data using SQL and Python, improving predictive power by 15% via vintage and roll-rate analysis. Trained challenger models (CART, Random Forest, XGBoost) integrated into the legacy enterprise system for a 10% relative Gini improvement.