Hriday Saha
OPEN TO WORK
HRIDAY SAHA · SENIOR DATA SCIENTIST · ML · NLP · LLMS/RAG
OPEN TO GLOBAL RELOCATION · VISA SPONSORSHIP WELCOME · AVAILABLE NOW
From Questions to Clarity,
One Dataset at a Time
Behind every chart is a choice. I turn noise into narratives — connecting algorithms to impact, and telling stories that move both people and metrics.
$1.5M+
FRAUD PREVENTED
Real-time scoring on 1M+ transactions a day, 94.2% F1.
96%
AUDIT EFFORT REMOVED
NLP control mapping at AXA — roughly $180K saved a year.
150+
PRACTITIONERS TRAINED
Mid-career professionals into job-ready ML, 92% completion.
4+
YEARS IN PRODUCTION
Insurance, banking and compliance, across three countries.
FLAGSHIP PROJECT
WORK
2023
Fraud Detection Platform
End-to-end ML platform that detects and prevents fraudulent transactions in real time.
Python
XGBoost
FastAPI
Spark
PostgreSQL
SHAP
$1.5M+
PREVENTED ANNUALLY
94.2%
F1 SCORE
<100ms
INFERENCE LATENCY
70%+
FEWER MANUAL REVIEWS
Reproduced with synthetic data due to employer confidentiality. Architecture, lifecycle design and serving performance match the deployed production system.
WORK
Automating Security
Framework Mapping
NLP system that maps audit findings to compliance controls at 93% Top-1 accuracy.
Python
Sentence-BERT
BM25
LightGBM
PERSONAL
FinSight: Leakage-safe financial forecasting
Leakage-safe financial forecasting, built point-in-time so backtests mean something.
Python
XGBoost
LightGBM
Prophet
PERSONAL
ClauseCheck: Insurance Policy Q&A (RAG)
Clause-grounded RAG that answers insurance questions and cites the policy line it used.
Python
RAG
FAISS
BM25
PERSONAL
Drift-watch: ML Model Monitoring Service
Model monitoring service for data drift, performance decay and alerting.
Python
FastAPI
PostgreSQL
Docker
Data isn't just numbers or algorithms. It's a vast, untapped library of human stories. Every dataset carries a record of real behaviour, real decisions, real trade-offs. My job is to find that record and turn it into something people can actually use.
Over 4+ years building AI and ML solutions for global enterprises, I've helped prevent $1.5M+ in fraud, automated high-volume operations, and pushed teams toward smarter decisions. Good analysis gets you the answer. Clear storytelling gets people to act on it.
For me, data science was never only about the model. It's about what the model means and what people do with it. I don't stop at analysis. I carry it through to something businesses, their systems, and the people behind them can actually use.
What I Love Solving
Six problems I keep coming back to. Each one is a thing that went wrong somewhere real, and the number underneath is what changed when it got fixed.
Catching it before the money leaves
Fraud that clears in 80 milliseconds can't be caught by a report on Monday. Velocity features, geo-mismatch, device fingerprints, merchant-to-user graphs — scored inline, under the latency budget.
$1.5M+
PREVENTED A YEAR
Reading what nobody has time to read
Thousands of audit findings, one framework, and a human matching them by hand every quarter. Retrieval plus a re-ranker turns 384 hours into an afternoon of review.
93%
TOP-1 ACCURACY
Forecasts that survive the real world
Most backtests are quietly cheating — the training set knows things it couldn't have known yet. Point-in-time construction first, model second. Sometimes the honest answer is that the baseline wins.
0
LEAKED FEATURES
Spending where it actually changes something
A model that predicts who will convert is not a model that tells you who to spend on. Uplift separates the persuadable from the already-decided, then turns that into a budget policy.
+0.0905
QINI UPLIFT
Knowing when a live model has gone stale
Accuracy on launch day is the easy part. Population shifts, upstream schemas change, and nothing errors — the model just quietly gets worse. PSI-based drift checks and alerts that fire before a stakeholder notices.
1M+
DAILY SCORES MONITORED
Answers people are willing to act on
An auditor won't accept a score with no reason attached. Confidence bands, key-phrase highlights, SHAP contributions — the output has to argue for itself before anyone changes a decision because of it.
40+
STAKEHOLDERS ADOPTED
the Man behind the Models
Hi! I'm Hriday, a data scientist by craft and a storyteller by instinct.
My journey into data didn't begin with metrics. It began with questions, the kind that sit with you over a quiet cup of tea. I've always been drawn to the why behind things: why people behave the way they do, why patterns repeat, why signals hide in noise. Somewhere between algorithms and intuition, I found my place.
Over the years, I've built systems that understand human language, detect fraud in milliseconds, automate repetitive workflows, and bring models into production with elegance and intent. But to me, it's never just about building. It's about building well. I believe in clean code, thoughtful design, and finding beauty in the details.
Outside of work, I live a little in stories. I love movies, anime, and books, and yes, I like my pages crisp and uncreased. I'm a tea person, the kind who slows down for the first sip. I like things tidy, precise, and purposeful, whether it's a script or a shelf. Detail matters to me. It always has.
For many years, in life and in code, I had a quiet companion: Shiro, the co-pilot, my golden boy. He was always nearby, often curled up under the desk while I debugged, designed, and discovered. He never debugged a line of code. But somehow, he debugged me. He always knew when to sit a little closer, or when it was time to step away from the screen. He passed away last year, but his calm presence still echoes in my work — in the pauses, the patience, and the quiet pride in doing meaningful things well.
This space, this story, carries a piece of him. Always.
Meet My Current Team
Teddy
DIRECTOR OF DRAMATIC AFFAIRS
Moody, independent, dramatic, but loyal to those she loves deeply
Blue
TERRITORY SECURITY MANAGER
Always alert, fiercely loyal, guards everything that matters most.
Choco
CHIEF HAPPINESS OFFICER
Spreads joy, wags tails, and lifts spirits wherever he goes.