Skip to content

Resume

Nripesh Pradhan

Head of Risk Intelligence & Automations · Chipper Cash · San Francisco Bay Area

Machine learning engineer with 7+ years designing and shipping production ML and LLM systems for a 10M+ user financial platform across 7+ global markets. I built the risk and compliance stack from scratch, real-time scoring services, LLM decision pipelines with rigorous evals, anomaly detection, and the data infrastructure underneath, operating at staff scope leading a quantitative engineering team.

7+years
building production ML and data systems
10M+users
served by systems I designed and shipped
$1.4M+/yr
in fraud losses prevented by one ML model
6,000+/day
fraudulent signups blocked in production

Selected impact

Shipped an LLM decision system with real evals

A two-model LLM pipeline that adjudicates watchlist screening alerts, validated against a double-blind golden dataset before launch, then run over a 148K-alert backlog at four cents a decision with precision and recall tracked in production.

Read the case study

Built real-time ML scoring services

A CatBoost model that scores every new user within an hour of signup, blocking 6,000+ fraudulent referrals a day, and a card-deposit scoring service that runs velocity and rolling-aggregate features at sub-5s latency.

Read the case study

Automated a fraud desk with LLMs

An autonomous pipeline that ingests unstructured partner reports, applies LLM extraction and identity resolution, and executes account actions in under 20 seconds, down from 30 minutes of manual work. Runs 24/7.

Read the case study

Rebuilt the data platform

Led company-wide dbt adoption, designed the canonical financial data models 15+ engineers build on, replaced a managed runtime with a self-hosted one, and shipped the monitoring that catches silent failures.

Read the case study

Experience

Oct 2025 - Present

Head of Risk Intelligence & Automations

Chipper Cash

Lead a quantitative engineering team building ML systems, real-time pipelines, and automated decision engines across 7+ global markets.

  • Drove the multi-year systems strategy, begun as tech lead, of replacing third-party vendor platforms with purpose-built quantitative systems, cutting $500K+ a year in costs and reducing end-to-end decision latency by over 80%.
  • Designed and built a high-throughput processing platform from scratch in Python and PostgreSQL, orchestrating 53,000+ events, 20,000+ investigation workflows, and 900+ automated filings across 8 currencies with deterministic audit trails.
  • Architected a real-time signal-detection pipeline evaluating millions of financial transactions against 20+ configurable statistical rule sets, surfacing 2,000+ actionable signals daily at sub-5-minute latency.
  • Built a dual-model NLP system for entity resolution and probabilistic match scoring across millions of records, achieving 85% precision and 95% recall against custom-curated international datasets.
  • Engineered an autonomous pipeline that ingests unstructured reports via API, applies LLM-based extraction and identity resolution, and executes automated actions in under 20 seconds, down from 30 minutes.
ML systemsLLM engineeringTeam leadershipCase study

Jan 2022 - Oct 2025

Staff Tech Lead, Risk & Growth

Chipper Cash

Promoted to Staff in role. Designed and shipped ML-powered scoring systems and production services handling real-time financial data.

  • Designed and deployed a real-time ML scoring model combining IP geolocation, behavioral timing, device fingerprinting, and identity-graph signals. Blocked 6,000+ bad actors a day, saving $1.4M+ a year.
  • Built a real-time ML scoring service for every card deposit, combining velocity features, rolling statistical aggregates, and behavioral signals at sub-5s latency. Intercepts $20K+ a day in losses.
  • Deployed a real-time NLP classification service that parses unstructured transaction data, extracting merchant identity, category, and metadata with sub-second latency at production scale.
  • Reduced account-takeover events from 30+ a day to under 10 a quarter with a multi-signal authentication system combining biometric verification, device reputation scoring, and behavioral anomaly detection.
  • Automated a dispute engine applying rule-based qualification, generating structured evidence packages, and submitting via processor API. Recovered $100K+ in disputed funds.
Real-time MLProduction servicesFraud detectionCase study

Feb 2021 - Jan 2022

Senior Data Scientist

Chipper Cash

Built core financial data infrastructure, statistical analysis frameworks, and production monitoring systems.

  • Led company-wide dbt adoption and designed the foundational financial data models powering all analytics, dashboards, and internal tools. Established SQL standards and code review across a 15+ person engineering team.
  • Conducted quantitative investigations that shaped policy: surfaced referral-abuse patterns during a 50K-a-day user spike, analyzed loss distributions by typology, and identified anomalous return rates that led to new automated controls.
  • Built a self-service experimentation platform enabling targeted launches reaching 900K+ users, and production alerting monitoring transaction rates across all payment providers in real time.
Data platformdbtExperimentationCase study

Oct 2018 - Feb 2021

Senior Consultant, Data Science

Logic (acquired by Accenture)

Client-facing quantitative consulting for Fortune 500 retail.

  • Delivered predictive modeling and quantitative analytics for Best Buy and Kendra Scott, leading engagements end to end: demand forecasting, inventory optimization, and pricing strategy, presented to senior leadership.
ConsultingForecasting

Systems shipped

Real-time ML & scoring

User Onboarding Score

CatBoost; IP, device, and identity-graph signals; 6,000+ blocks a day

Card-Deposit Scoring Service

velocity and rolling aggregates at sub-5s latency

Account-Takeover Defense

multi-signal authentication; 30+ a day to under 10 a quarter

Merchant Parsing Service

NLP classification of raw transaction strings, sub-second

LLM systems

Watchlist Decision Support

two-model LLM pipeline with golden-dataset evals

Fraud-Desk Automation

LLM extraction and identity resolution; 30 minutes to 20 seconds

LLM Monitoring Framework

six standardized signal types from one YAML config

Data platform

dbt Models

the canonical financial models every downstream system is built on

Transaction Signal Engine

20+ statistical rule sets; 2,000+ signals a day at sub-5-minute latency

Data Monitoring System

config-driven checks with auto-segmented root cause

Platforms & growth

Case Processing Platform

53K+ events and 900+ automated filings with deterministic audit trails

Campaign Delivery Engine

Python + Dask; 900K users per run

Risk-Aware Payout Pipeline

re-checks risk flags at the moment money moves

Toolbox

Languages

Python · SQL · TypeScript

ML / AI

LLMs · OpenAI API · CatBoost · scikit-learn · NLP

Data systems

dbt · Snowflake · PostgreSQL · Airflow · Pandas · NumPy

Infrastructure

Docker · GCP · AWS · Flask · Neo4j · Pinecone · Datadog

Education

Grinnell College

B.A. in Mathematics & Statistics

2018

Budapest Semesters in Mathematics

Study abroad in advanced mathematics

Stanford Continuing Studies

Coursework in literature and finance

HBX CORe, Harvard Business School

Business analytics and economics

Beyond the systems work, I write: essays on fintech and risk, the craft of building, and the books I keep turning over.