Senior machine learning engineer with 5+ years in industry, 3+ building production ranking systems that drive revenue. I turn complex business challenges into ML solutions with real impact.
I am a Senior Associate Algorithm Engineer at QuinStreet, where I own the end-to-end machine learning ranking architecture for the Personal Loans vertical, spanning CTR and conversion modeling, optimal allocation, constrained marketplace optimization, and online experimentation. Across 5+ years in AdTech and data engineering, I’ve focused on production ML systems where modeling decisions move revenue. Day to day, my work centers on matching consumer intent to the right offers in real time and guarding results with experimentation and monitoring. I also set modeling standards through design reviews, code reviews, and mentoring.
I hold a Master’s in Electrical and Computer Engineering from UCLA (GPA 3.97/4.0), where my research on equitable remote health sensing was published at SIGGRAPH 2022 and I co-authored two chapters (Introduction to Computational Imaging and Computational Toolkit) for the MIT Press Computational Imaging textbook. Earlier, I spent two years as a Software Engineer at Accenture in India, building data solutions for the healthcare and pharma sectors, and interned in ML at VidMob.
I’m passionate about using technology to drive meaningful impact, and I’m looking to tackle high-leverage ranking, ads, recommendation, and marketplace optimization problems at scale. Outside work, I pursue Bharatanatyam, an Indian classical dance form, and cooking, both of which keep me focused and composed.
Senior Associate Algorithm Engineer · Personal Loans Ranking
July 2026 – Present
Technical owner of the end-to-end ML ranking architecture for the Personal Loans vertical across 3M+ annual inquiries; scope expanded in 2026 to include the conversion model alongside CTR models, owning ranking performance, deployment strategy, experimentation, and expansion from subprime to higher-quality traffic (~$2M margin opportunity). Set the roadmap with Product, Business, and Engineering.
Built the Personal Loans XGBoost vendor conversion (CVR) model and own the Personal Loans CTR model with weekly cron retraining; a persistent 90/10 holdout against the legacy lookup baseline shows about a 10% profit lift on a $1M+ margin segment.
Lead a margin-preservation initiative (currently in pilot) combining CTR and CVR predictions to filter unprofitable leads before purchase, projecting an 8–12% reduction in wasted lead spend.
Onboard and mentor 4+ algorithm engineers on ranking architecture, design reviews, and code reviews; authored the team’s code-quality and model-release standards, now followed team-wide.
Built and shipped XGBoost CTR ranking models for Banking, Auto Insurance, and Personal Loans, averaging a 10% improvement over legacy 14-week lookup-table baselines in live serving.
Built and productionized the Personal Loans funnel conversion prediction model using Bayesian optimization and cross-validation, complementing CTR models to optimize downstream conversion.
Implemented the Hungarian algorithm for optimal constrained ranking assignment and validated it against the greedy baseline through staged online A/B tests (staged 90/10, then 50/50) on matched publisher traffic, lifting margin revenue 2.5–8% depending on traffic; now live in Auto Insurance and Banking.
Built and optimized an automated simulator that replays ranking changes against a replica of the live system, so the team can test changes safely before release. Cut its runtime by 50% while keeping its behavior matched to live production logic. It also runs daily to monitor Auto Insurance production health.
Engineered high-impact user and lead features using domain-specific aggregations, temporal windows, statistical summaries, and one-hot encoding, guided by SHAP attribution, permutation importance, and correlation analysis; built Amazon Redshift and Tableau evaluation workflows and partnered with Platform Engineering to validate and deploy models.
Built deep learning models on the Creative Analytics team in Python and PyTorch, including celebrity detection and brand audio mention models, to score creative ad assets and support automated engagement prediction and recommendation pipelines.
Designed and implemented computer vision models for object detection and image scoring, lifting ad engagement classification accuracy by 25%.
Researched new methods to give clients data-driven guidance on boosting ad engagement, and benchmarked user interaction-rate models via A/B testing.
Engineered modular, reusable components for data preprocessing, model training, and evaluation, reducing pipeline execution time across ML iterations.
Software Engineer · Healthcare & Pharma Data Engineering
June 2019 – July 2021
Automated pharma sales ETL workflows in Python, SQL, and Apache Hive, cutting analysis time by 50%, and automated recurring support-team processes.
Built a customer-facing interface for the client, drug-sales analytics, and Tableau dashboards that lifted competitor match rates by 20%, and delivered weekly data trends for the pharma client recognized by the Australian team.
Trained 10 new graduate engineers and was named an Accenture Global Innovation Challenge Semifinalist.
July 2020 – January 2023 · Los Angeles, California · Advised by Prof. Achuta Kadambi
Co-authored 2 chapters (Introduction to Computational Imaging and Computational Toolkit) for the textbook Computational Imaging published by MIT Press.
With a fellow researcher, derived a shift-robust loss for rPPG that learns from misaligned camera and pulse-oximeter data, decreasing error by 40%.
Research Intern · Artificial Intelligence and Robotics Lab, Indian Institute of Science (IISc)
January 2019 – June 2019 · Bengaluru, India · Advised by Dr. Suresh Sundaram
Cleaned and rearranged the ABIDE fMRI dataset based on correlation between brain regions to pioneer faster and more accurate Autism diagnosis.
Compared several CNN architectures on the cleaned data to find which performed best, developing deep neural networks that achieved 80% accuracy (GitHub · Report).
Teaching Associate · University of California, Los Angeles
March 2022 – June 2023 · Los Angeles, California
Served as Teaching Associate for two courses: Mathematics for Life Scientists (Life Sciences Dept.) and Food Politics in the World of Arts and Culture (Dance Dept.).
Led 2 discussion sections and 2 lab sessions guiding students in mathematical modeling in Python, prepared course materials for BruinLearn and Zoom, and supported faculty with grading and student feedback.
Track: Signals and Systems track, focused on computer vision and machine learning.
Coursework: Computer Vision, Large Scale Data Mining, Large Scale Social and Complex Networks, Neural Networks and Deep Learning, Machine Learning and Modeling in Bioengineering, Computational Robotics, Computational Imaging, Decision Making in Stochastic Processes, Technology Management, Physics and Informatics of Medical Imaging.
PES University, BengaluruB.Tech. in Electronics and Communication Engineering · Aug 2015 – Aug 2019 · CGPA 8.5/10
Honors & Certifications: CNR Rao Merit Scholarship (Semester 7), Distinction Award (Semesters 1, 5, 6, and 7), German A1 Certification (Dec 2017).
Coursework: Introduction to Python, C & Embedded Systems, Probability and Random Processes, Linear Algebra, Information Theory and Coding, Machine Learning, Digital Signal Processing, Speech Processing.
Adversarial multimodal fusion of 77 GHz millimeter-wave radar and RGB video to mitigate skin-tone bias in remote heart-rate estimation, improving accuracy across diverse skin tones by 75% (SIGGRAPH Paper · EquiPleth Code).
Co-authored Introduction to Computational Imaging and Computational Toolkit with the UCLA Visual Machines Group, covering foundational optics, signal processing, and computational imaging systems (MIT Press Book).
eYantra 2016 National Robotics Competition: Coached a team of 3 to the national semi-finals.
Epsilon 2016 Operations Lead: Led operations across 100+ events and 2,000+ participants.
IEEE SSCI Bangalore: Presented research work at the IEEE Symposium Series on Computational Intelligence.
Centre for Intelligent Systems: Active student researcher and member.
Bharatanatyam: Professional Indian classical Bharatanatyam dancer.
04. Selected work
Case studies and projects across marketplace ML ranking, constrained optimization, computer vision, and embedded systems. Browse the archive for additional details.
Personal Loans ML ranking & margin preservation
Owning the end-to-end CTR, conversion (CVR), and lead-filtering ML ranking architecture across 3M+ annual Personal Loans inquiries at QuinStreet.
The challenge
Legacy lookup-table baselines struggled to capture borrower intent and publisher traffic quality across 3M+ annual inquiries, leaving margin untapped and spending on low-conversion leads.
My contribution
Built the Personal Loans XGBoost vendor conversion model and own the CTR model with weekly cron retraining, SHAP-guided feature engineering, and a joint CTR and CVR lead-filtering system (in pilot).
XGBoostCTR & CVRSHAPRanking
Hungarian vs. Greedy Constrained Marketplace Optimization
Replacing greedy offer slotting with global optimal constrained assignment (Hungarian algorithm) validated across staged online A/B rollouts.
Problem & approach
Problem: Greedy offer allocation sub-optimally matched consumer inquiries under advertiser eligibility and budget constraints. Approach: Formulated slotting as a global constrained assignment problem using the Hungarian algorithm to maximize expected marketplace margin across eligible offers.
Validation & next steps
Validation: Replayed against a replica of the live system before release and validated online through staged A/B rollouts (staged 90/10, then 50/50) against the greedy baseline on matched publisher traffic. Next steps: Extending constrained assignment with joint CTR × CVR expected-margin scoring across additional verticals.
Blending RGB camera video with 77 GHz millimeter-wave radar and a shift-robust loss function to mitigate skin-tone bias in remote heart-rate sensing.
The challenge
Camera-based remote photoplethysmography (rPPG) degrades on darker skin tones due to optical absorption, and unsynchronized camera-oximeter capture introduces temporal lag.
My contribution
Contributed to an adversarial multimodal fusion pipeline combining 77 GHz radar and RGB video at UCLA VMG, and with a fellow researcher derived a shift-robust PyTorch loss function for temporally misaligned PPG signals.
Restructured the ABIDE neuroimaging dataset using inter-regional brain correlation analysis and compared multiple CNN architectures on the cleaned data, achieving 80% early-detection accuracy.
With a fellow researcher, derived a shift-robust regression loss function for remote photoplethysmography (rPPG) that learns directly from temporally misaligned RGB camera and pulse-oximeter signals, decreasing pulse prediction error by 40%.
University of California, Los Angeles · Jan 2023 – Mar 2023
Explored deep learning architectures, including CNNs, LSTMs, GRUs, RNNs, VAEs, Transformers, and attention mechanisms, for multi-class classification of EEG motor imagery signals, achieving 75% accuracy on the entire dataset across single-subject and full-dataset evaluations.
Engineered a Raspberry Pi robotic crane where a deep neural network classified camera images of waste and the crane placed items into the matching biodegradable, non-biodegradable, or electronic-waste bin with 95.18% accuracy (comparing PCA, LDA, and CNNs with transfer learning).
I’m always happy to talk machine learning ranking systems, ads, recommendation engines, or marketplace optimization at scale. The best way to reach me is through LinkedIn.
