REF SB / 2026 / AI-GENAI SEGMENT BFSI × AGENTIC AI LOCATION Bhubaneswar, Odisha, India STATUS OPEN TO RELOCATION

PGDBA, IIM Calcutta. B.Tech, NIT Trichy.
Now building agentic AI.

Sambit Behura is a data scientist and AI/GenAI engineer — a PGDBA from IIM Calcutta, ISI Kolkata & IIT Kharagpur, and a B.Tech from NIT Trichy, with 9+ years of professional experience, including 6.5+ years across McKinsey, American Express, and consumer lending. He built the scorecards and experiments that drove ~$200M in incremental business, and is now applying that same rigor to designing, evaluating, and deploying production multi-agent RAG systems.

APPROVED FOR AGENTIC
AI PRODUCTION
6.5+
Years — McKinsey, Amex, consumer lending
$200M
Annual incremental billed business, Amex
0.85
GINI — B2B recommendation model
2
Multi-agent RAG systems in production on AWS

Every role fed the next model.

Data Science Intern (Pre-Placement Offer)
McKinsey & Company Oct 2019 – Mar 2020

Built a personal-loan cross-sell model with XGBoost and Random Forest, targeting early-tenure consumer-durable and two-wheeler loan customers.

Data Scientist
U GRO Capital Apr 2020 – Nov 2021

Built U GRO's first-generation business-loan scorecard from scratch — 1,000+ engineered features from bank statement and bureau data, shipped as a logistic regression model (GINI ~0.80) that became the core of underwriting.

Senior Manager, Analytics
SMFG India Credit (Fullerton India) Dec 2021 – Jun 2022

Monitored and validated Application, Behavioural, and Collection credit risk models via GINI, PSI, and CSI stability metrics — catching drift and recommending recalibration.

Analyst → Assistant Manager, Global Credit Services
American Express Jun 2022 – Nov 2025

Moved from Client Onboarding & Digital Capabilities into the AI & Modeling team over 3.5 years. Built an XGBoost B2B recommendation model (GINI 0.85, 70+ features) and quantified ~$200M in incremental billed business through randomized test-control experimentation.

Senior Consultant
EXL Mar 2026 – Apr 2026

Built fraud detection models for a Long Term Care insurance client, flagging anomalous claim patterns from claims behavioral data.

Independent GenAI Build
Agentic AI Portfolio 2026 – present

Shifted from predictive scorecards to production agentic AI — designing, evaluating, and deploying multi-agent RAG systems on AWS, with the same rigor around GINI/PSI monitoring now applied to LLM-as-judge evaluation.

What's under the hood, grouped like a feature set.

GenAI / LLM

01
CrewAI orchestrationLangChainRAG architecturePineconeChromaDBHybrid retrieval (BM25 + dense + rerank)Structured outputsLLM-as-judge (RAG Triad)

ML / Data Science

02
PythonXGBoostRandom ForestLogistic RegressionSHAPFeature engineeringA/B testingGINI / PSI / CSI monitoringCredit scorecards

MLOps / Cloud

03
AWS ECS FargateECRS3CloudWatchDockerGitHub Actions CI/CDStreamlitSQL

Certifications

04
Building & Evaluating Advanced RAGLangChain for LLM App DevelopmentMachine Learning A-ZCredit Risk Modelling in Python

Built, deployed, and evaluated — not just prototyped.

Chargeback Intelligence Agent

Deployed

A compliance-grade RAG pipeline that grounds card-network chargeback defenses in the actual Visa/Mastercard rule text — not the model's memory of it.

Query expansion Hybrid retrieve (Pinecone + BM25 + rerank) Structured, validated verdict Audit log (S3)
0.97answer relevance
0.75groundedness
0.74context relevance
400stratified CFPB cases evaluated
github.com/sbehu/chargeback-intel-agent

AI Financial Analyst Platform

Deployed

A multi-agent workflow that reads six years of American Express 10-K filings and benchmarks them live against Visa and Mastercard — then checks its own arithmetic.

CrewAI multi-agent parse Year-isolated retrieval Dedicated math-verification agent Live competitor benchmarking
200+question golden eval set
6 yrsof 10-K filings analyzed
AWSECS Fargate + Docker + GitHub Actions
github.com/sbehu/financial_agent_intel

Multimodal Loan Document Verification Agent

In build

A GPT-4o vision agent that inspects loan documents the way an underwriter would — checking three distinct fraud-detection scenarios rather than a single classifier score.

GPT-4o vision CrewAI orchestration Pinecone + S3 Streamlit review UI

Open to BFSI, fintech, and AI-consulting roles.

Currently exploring GenAI/agentic AI engineering and data science roles at BFSI-focused GCCs, fintech product companies, and AI consulting firms — open to relocation across Bengaluru, Hyderabad, Gurugram, and Mumbai.