Full-stack data science, AI & machine learning

Data science that holds up in production.

Three years building credit-underwriting and fraud models at a lending fintech, plus independent ML work and a self-built computer-science education.

72%+AUC

Risk-model performance

16%

Cohort risk reduction

200M

SMS processed a day

450%+

Fraud-detection improvement

3

Data scientists led

3 yrs

At MoneyView, Apr 2021 – Jul 2024

At MoneyView

Underwriting at scale.

As Manager, Data Scientist L3 at MoneyView (Whizdm Innovations), I built the machine-learning models behind credit underwriting. I led three data scientists I had trained straight from college, and trained 15 fresh graduates in hands-on machine learning and Python for data science (rated 5 stars).

That meant a pipeline turning raw SMS data into features at 200 million SMS a day, a risk model that reached 72%+ AUC and cut risk in the targeted cohort by 16%, and a set of fraud-detection improvements that exceeded 450%.

Inside the pipeline, 50 TB of SMS data became 330+ features on lifestyle, financial literacy and cash flow. Target encoding and normalisation removed null values, outcomes were simulated for rejected applicants, and the model kept was the one that stayed stable across time and segment cohorts. An offline feature store on MySQL, run by cron jobs on EC2, served over 1 lakh users a day.

Beyond the first model, a bank-statement model carried an expected risk reduction of over 26%, and proofs of concept were approved for an alternate-data model and for a meta-model that simplifies reading several models’ predictions. Fraud work raised mobile-device detection by more than 450% through SMS fingerprinting, and a deep-learning model to find face photographs in selfies was under way. Flask, Plotly and Dash dashboards for credit, risk and collections teams saved each analyst two hours a day.

Day-to-day tools were Python, SQL, pandas, scikit-learn, Flask, Plotly, AWS and Metabase.

Independent work

Learning in public.

Outside employment, in 2020 and 2021, I built a fake-news classifier (TF-IDF with a passive-aggressive classifier), a spam classifier (Naive Bayes and XGBoost), a cat-image classifier (a multi-layer neural network), a stock-portfolio web app with live prices, and the first terminal version of the card game Declare, alongside GIS dashboards and automation bots for clients. Since 2019 I’ve followed a self-built curriculum drawn from MIT 6.0001 and 6.0002, Harvard CS50 and Stanford CS229. See the education page.

Today the same skills drive the analytics and AI systems that YourTechConsultants built for its client Puranmal Sons, from ERP-fed dashboards to a Gemini-assisted WhatsApp CRM, commissioned and approved by Puranmal Sons’ leadership.

Toolkit

Stack.

  • Python
  • SQL
  • pandas
  • NumPy
  • scikit-learn
  • Matplotlib
  • Flask
  • FastAPI
  • Plotly
  • Dash
  • Jupyter
  • Metabase
  • AWS S3
  • EC2
  • Athena
  • SageMaker
  • Glue
  • Postgres
  • SQL Server
  • Gemini
  • Claude

Questions

Data science, answered.

What does a data scientist do in credit underwriting?

They build models that estimate how likely a borrower is to repay and how likely an application is to be fraudulent, using applicant, credit-bureau and alternative data, so lenders can approve, price or decline faster and with less risk. At MoneyView, Raghav Palriwala built these models and the data pipelines that feed them.

What did Raghav Palriwala do at MoneyView?

From April 2021 to July 2024 he was Manager, Data Scientist L3. He led a three-person team, built ML models for credit underwriting, an SMS-data pipeline processing 200 million SMS a day, a risk model with 72%+ AUC that reduced cohort risk by 16%, and fraud-detection improvements exceeding 450%.

Which tools does he use for data science?

Python, SQL, pandas, scikit-learn, Flask, Plotly and Metabase for modelling and reporting, with AWS (S3, EC2, Athena, SageMaker, Glue) for data and deployment. For newer work, he builds LLM-integrated apps with Gemini and Claude.

Can I hire him for data science or machine-learning work?

Yes, remotely and worldwide, through YourTechConsultants. Dashboards, analytics pipelines, ML models and AI-integrated tools are all in scope. Email rmpalri@gmail.com with a short brief.

Have data and a question?

Let’s see whether it can be answered.