Work

10+ years of pricing risk with data

From forecasting and drug repurposing at Wipro, to credit and surety at Swiss Re, to leading AI for credit reinsurance at Marsh Re.

Experience

2025 to present
London

SVP, Credit Reinsurance AI

Marsh Re (formerly Guy Carpenter)

Leads data and AI for Credit, Bond and Political Risk, with a focus on actuarial pricing in credit reinsurance.
Builds on Databricks across pricing, risk and portfolio workflows.
Develops LLM products for pricing support, entity matching, data transformation, wording analytics with RAG, contract management, workboard automation, email analytics and bordereaux management.

Databricks
PySpark
LLMs
RAG
Agentic systems

2020 to 2025
London

Lead Data Scientist and Product Owner, Credit & Surety

Swiss Re

Owned the end-to-end data and AI landscape for Credit & Surety on Palantir Foundry.
Worked with Prof. Dimitris Bertsimas (MIT) on contract bond default prediction using optimal classification trees.
Built underwriting memo generation and document summarisation with generative AI.
Delivered £400k in IT cost savings through Palantir Foundry, and released 2 FTE by automating bordereaux reporting.

Palantir Foundry
Palantir AIP
PyTorch
FinBERT
Azure ML

2016 to 2020
Bangalore

Data Science Manager, Advanced AI Practice

Wipro

Built sales prediction models for major clients, including Levi’s forecasting work presented by the CFO at the World Economic Forum in Davos.
Led a COVID-19 drug repurposing AI programme in partnership with Stanford University, later continued by the Government of India.
Drove high-accuracy time-series forecasting for revenue and cost strategy.

PySpark
XGBoost
AWS SageMaker
GCP
IBM Watson

Independent work

Projects

Apps built outside the day job: a creative studio for seeing risk differently, a private record-matching tool, and pricing engines applying hazard, vulnerability and portfolio-capital thinking to different kinds of risk.

Live Creative work

Creative Risk Studio

Risk isn’t a number, it’s a chain reaction. A growing collection of experiments for seeing it that way: break a country on the Contagion Globe and watch credit risk travel along real trade routes, play your way through the maths, replay a crisis, or simply ask the studio’s AI a question out loud.

Svelte
TypeScript
Vite
D3 globe
Live public data

Open the studio

Personal project, built in my own time. Uses public data such as USGS, GDACS, NASA, the World Bank, FRED and UN Comtrade; shock probabilities and pricing loads are illustrative. Not affiliated with my employer.

Contagion Globe: pick a shock, such as sudden sanctions or a sovereign default, and see receivables at risk and expected credit-insurance claims ripple along real trade links
Games that teach: a Quant Arcade from gambler’s ruin to Black-Scholes, Merton and Markov rating migration, plus role-play games like Be the Reinsurer and Renewal Day
A quant course from roulette to Markowitz, and an efficient frontier for a reinsurer’s lines of business
Crisis replays and cascades: Suez, COVID and 2008, shock chains with probability, delay and cost on every link
Ask Banu, a spoken AI assistant, plus live hazards, credit markets and Voyager 2 tracked live from NASA

Live Private by design

MatchFlow

Company name matching, done properly. Make two messy lists agree, and see why: MatchFlow links records across systems when names, spellings and formats disagree, such as “Veloria Steel Hldgs Ltd” in a CRM and “VELORIA STEEL HOLDINGS LIMITED” in an ERP. Clear matches link themselves, uncertain ones queue for a person with the evidence laid out, and nothing is ever uploaded.

React
TypeScript
Web Workers
Vite
FastAPI

Open MatchFlow

Personal project, built in my own time on published record-linkage methods. Free and open for lists up to 500 rows; the sample data is synthetic. Not affiliated with my employer.

Two modes: match list A to list B into a reviewed crosswalk, or standardise and deduplicate a single list
Explainable: per-field evidence, Fellegi-Sunter style weights, named penalties for contradictions such as a different country or “Fund II vs Fund III”, and the margin over the runner-up
Private: files are read and matched in Web Workers in your browser, so no row ever reaches a server
Measured: on its synthetic answer key, 251 of 251 automatic links are correct and every true match is surfaced
Built to scale: blocking scores about 0.7% of the 26 million possible pairs in a 5,000 by 5,000 benchmark

Live

Credit & Surety Pricing Engine

Prices credit, political risk, surety and SRT on one structural engine. Treaty and facultative, whole books and single names, bonds and tranches, with every price explained and traced back to the same simulated economy.

Python
FastAPI
NumPy & SciPy
React
TypeScript

Open the app

Personal project, built in my own time on open, published methods. Synthetic books and fictional names only. Not affiliated with my employer.

Treaty pricing for quota share, excess of loss and stop loss, stacked or side by side, with reinstatements
Single risk credit and surety bonds, priced on their own and against the book
SRT tranche pricing, with the investor and bank views moving together
Merton asset model with seeded, correlated economies, 10,000 simulated years and 1-in-200 tail capital
Claim simulator, cash flows, price adequacy and a built-in pricing assistant

Live · proof of concept First of its kind

Influencer Risk Engine

The first quantitative pricing engine for creator income risk. Deplatforming, algorithm shocks and scandals can wipe out a creator’s income overnight, yet no insurer prices that risk quantitatively today. This engine does, by borrowing the machinery of catastrophe reinsurance.

Python
FastAPI
NumPy & SciPy
GARCH
Regime switching
React

Open the app

Personal proof of concept, built in my own time on open, published methods. Illustrative calibration, labelled in the app. Not affiliated with my employer.

Hazard layer: catastrophe-style modelling of platform algorithm changes, policy shifts and ad-market shocks, from a maintained event catalogue
Vulnerability layer: a per-creator exposure curve from alternative data, such as platform concentration, niche, audience quality and revenue diversification
Idiosyncratic layer: real-time NLP on sentiment and sentiment velocity for scandal and personal-event shocks
Portfolio and capital: Monte Carlo loss distributions, expected loss, EVT-fitted VaR and CVaR, economic capital and RAROC
Optionality: Merton jump-diffusion for pricing growth-linked structures

Skills

Risk & pricing

Credit risk
PD / LGD / EAD
Basel III
Reinsurance pricing
Credit & surety
Political risk
Portfolio optimisation
Stochastic modelling

AI & modelling

Generative AI
LLM systems
RAG
Agentic workflows
NLP
PyTorch
TensorFlow
Bayesian methods
Time series

Platforms & code

Python
PySpark
SQL
TypeScript
Databricks
Palantir Foundry
Azure ML
AWS SageMaker
FastAPI
Django

Recognition

NASSCOM AI GameChanger AwardCOVID-19 drug repurposing research
Best Employee of the YearSwiss Re
Pinnacle AwardRecognition
Frontline Manager AwardRecognition
Guest lectureSCMHRD, Symbiosis, Pune, 2022

Education

B.Tech, Civil EngineeringNIT Warangal, 2012 to 2016
First Class with DistinctionMerit Scholar

Want to talk about a role or a project?

The quickest way to reach me is email. I reply to every serious message.

Email me LinkedIn