RUSHIKESH HIRAY

Senior Data Scientist & ML Engineer

I design production ML systems for acquisition, segmentation, computer vision, GenAI, and optimization, turning complex data into decisions teams can trust and use at scale.

About

Production-grade machine learning systems for scoring, segmentation, detection, retrieval, and optimization.

Featured Client Work

HDFC BankML-Driven Loan Acquisition Engine
HDFC BankCustomer Risk Categorization and Segmentation
AECOMPipe Insight Defect Detection

Banking & Risk

Computer Vision

Autonomous Driving

GenAI & LLMs

Optimization

8+
Years Shipping ML
2 Cr+
Customers Processed
75%+
Top-Decile Capture
5
Enterprise Clients

Experience

Career progression and impact

Sr Lead Data Scientist

HDFC Bank

Apr 2025 - PresentMumbai, India
  • Built end-to-end Business Loan and Personal Loan acquisition models using XGBoost and CatBoost.
  • Engineered 300+ features across CASA, Credit Card, Digital Funnel, and Loan One-View datasets.
  • Scaled PySpark and SQL pipelines for 2 crore+ customers across 18 source systems and leadership-facing dashboards.
  • Improved top-2 decile capture from roughly 60% to 75%+ for acquisition campaigns.
XGBoostCatBoostPySparkSQLAutoencodersKMeans

Senior Data Scientist

iLink Digital

Nov 2022 - Apr 2025India
  • Built the Pipe Insight defect detection pipeline for AECOM using Faster RCNN-based inspection models.
  • Developed a RAG-powered financial document assistant for Abdiel Capital Advisors, LP.
  • Improved annotation tooling and OCR post-processing workflows to raise data quality and downstream model usefulness.
  • Reduced manual inspection effort for infrastructure workflows through automated defect detection.
Faster RCNNRAGLLMsOCRDeep LearningPython

ML Engineer - Computer Vision

Tata Consultancy Services

Jun 2020 - Oct 2022Pune, India
  • Built YOLOv4 object detection pipelines and DeepSORT tracking for real-time driving scenes.
  • Developed traffic sign classification and DeepLabV3 lane extraction models in TensorFlow.
  • Created internal annotation tooling in PyQt5 for image and LiDAR labeling workflows.
  • Reached 92 mAP at 0.7 IOU for object detection in autonomous driving use cases.
YOLOv4DeepSORTTensorFlowDeepLabV3PyQt5LiDAR

Optimisation Engineer

Tata Consultancy Services

Nov 2018 - May 2020Pune, India
  • Built a vehicle testing scheduler for Nissan using Microsoft Z3 SAT Solver and constraint programming.
  • Modeled vehicle, test, and dependency rules directly into mathematical decision logic.
  • Combined exact constraints with greedy optimization for practical schedule generation speed.
  • Reduced testing schedule implementation cost by 40%.
Z3 SAT SolverPythonConstraint ProgrammingGreedy Algorithms

Selected Projects

Technical depth and real-world applications

XGBoostCatBoostPySparkSQL

ML-Driven Loan Acquisition Engine

An acquisition decision engine for Business Loan and Personal Loan campaigns targeting existing-to-bank customers.

Approach

Framed campaign prioritization as a supervised ranking and classification problem using XGBoost and CatBoost.

Impact

Improved top-2 decile capture from roughly 60% to 75%+.

XGBoostCatBoostPySparkSQLFeature Engineering
AutoencodersKMeansPySparkDeep Learning

Customer Risk Categorization and Segmentation

A customer segmentation and recommendation foundation used to categorize risk and improve product relevance at scale.

Approach

Built Autoencoder plus KMeans segmentation over 150+ variables to uncover actionable customer clusters.

Impact

Improved Business Loan recommendation lift by 8-10x over business-as-usual baselines.

AutoencodersKMeansPySparkDeep LearningSegmentation
Faster RCNNPythonDeep LearningOCR

Pipe Insight Defect Detection

An automated inspection pipeline for identifying pipeline defects from imagery instead of relying only on manual review.

Approach

Built Faster RCNN-based detection models for defect localization and classification.

Impact

Reduced manual inspection effort by automating defect identification at scale.

Faster RCNNPythonDeep LearningOCR
YOLOv4DeepSORTDeepLabV3TensorFlow

Autonomous Driving Perception Stack

A perception workflow combining object detection, tracking, and lane segmentation for autonomous driving scenarios.

Approach

Built YOLOv4 object detection models for dynamic road scenes.

Impact

Achieved 92 mAP at 0.7 IOU for object detection.

YOLOv4DeepSORTDeepLabV3TensorFlowPyQt5
Z3 SAT SolverPythonConstraint ProgrammingGreedy Algorithms

Vehicle Testing Schedule Optimizer

A scheduling engine that transformed complex prototype vehicle testing constraints into faster planning decisions.

Approach

Modeled tests, vehicles, and dependencies as constraint satisfaction problems using Microsoft Z3 SAT Solver.

Impact

Reduced implementation cost by 40%.

Z3 SAT SolverPythonConstraint ProgrammingGreedy Algorithms
RAGLLMsPythonVector DB

RAG Assistant for Financial Research

A retrieval-augmented assistant that let analysts ask natural-language questions over financial documents.

Approach

Built a document retrieval and LLM answering workflow over financial research materials.

Impact

Accelerated research workflows with conversational access to financial documents.

RAGLLMsPythonVector DBNLP

Skills & Stack

Technical breadth across ML, engineering, and systems

Predictive Modeling

Build ranking and classification systems that improve targeting, acquisition, and customer prioritization.

XGBoostCatBoostScikit-learnModel Evaluation

Data and Feature Engineering

Turn fragmented source systems into stable, decision-ready feature pipelines for large-scale ML programs.

PySparkSQLFeature EngineeringETLPandas

Computer Vision

Deliver perception systems for inspection and autonomous-driving style environments where model quality must survive messy real data.

YOLOv4Faster RCNNDeepSORTDeepLabV3OCR

GenAI and RAG

Create retrieval-grounded AI workflows that help users interrogate documents and domain knowledge safely and efficiently.

RAGLLMsLangChainNLPVector Databases

Optimization and Decision Systems

Model operational constraints directly so the system recommends better schedules, allocations, and actions.

Z3 SAT SolverConstraint ProgrammingMathematical ModelingGreedy Algorithms

Education

Academic foundation and credentials

University of Pune

Bachelor's DegreeComputer Science

2014 - 2018

Certifications & Achievements

  • Deep Learning Specialization
  • Professional Scrum Master
  • Best Innovation Award - National Level Competition
  • Smart India Hackathon Participant

GitHub

Open source work and technical portfolio

@rhiray1996

Engineering and public-build proof of ML work.

Actively building new projects in GenAI, RAG Systems, and ML Pipelines.

Visit GitHub Profile

Resume

Download or preview my full resume

Rushikesh Hiray — Resume

A concise overview of my experience, skills, and background in machine learning and data science.

Download Resume

Get in Touch

Open to opportunities, collaborations, and conversations