Srinath Narayanan Email & Phone Number

Mountain View, Ca, Us

As part of the Retail Ads ShopEx effort, I work towards maximizing the usefulness of shopping Ads and show the best Ad suggestions on Product Ad Listings (PLAs) using ML models and LLM based architectures.

Mountain View, Ca, Us

As part of the [AdManager](https://admanager.google.com/) Sell-Side Infrastructure Quality team, I work towards increasing the sustainability of the free and open Internet, by optimizing publishers’ revenue from display ads platforms and exchanges.

New York, Ny, Us

Building world-class state of the art fraud detection and risk aversion models in Core Risk & Fraud Detection Quantitative Modeling Team in Customer and Consumer Banking line of business. I use multi-pronged approaches covering General AI, Deep Learning using Image processing, Sequence Modeling and Natural Language Processing in an efficient way using parallel and distributed computing.Inclearing check fraud & Digital Fraud models (In Production) : - Conceived and implemented a XgBoost based unified fraud detection framework for risk assessment of Chase's different monetary instruments. It handles nearly 100mm transactions a month, and captures around $20mm/month fraudulent items with a 86.5% detection rate.- Adapted an object extractor network using YOLOv3 to identify and crop check signatures, and textual information for parsing.- Designed and implemented a high accuracy Handwriting and Printed text Parser framework based on Convolutional Recurrent neural networks (CRNNs).- Contrastive learning of linear embeddings of signature crops using a Siamese network in an efficient manner for signature comparison.- Designed and constructed a robust testing framework, leading to faster prototyping using GPUs in production.- Using PySpark to model, clean, pipeline, analyze terabytes of transactional time-series data to infer cross-channel information to build first-of-a-kind machine learning transactional model using graph databases to model user interactions across channels.Other works:- Implementing a sequence to sequence embedding of user behavior to a denser latent space to identify fraudulent patterns.- Building a social graph of Chase's digital users using TigerGraph that leverage huge transactional and user repository for identifying fraudulent patterns.

La Jolla, Ca, Us

Course – CSE 258, Recommender systems and web mining. Responsible for setting and grading assignments, answering queries and holding office hours.

Franklin Lakes, New Jersey, Us

Developed predictive models for cost signal behavior, shortage forecasting and drug monitoring. Adapted domain knowledge to build active-intelligent inventory and budget monitoring systems using models built using R and Python.Drug budget management - • Wrote a R library for optimizing pricing strategies for pharma drugs, and forecasting demand-supply variations.• Satisfied business success criteria by achieving a 0.92 correlation in a 3-month window with 11% mean absolute percentage error (MAPE), by developing gradient boosting, time series LSTM and ARIMA models.• Led an intern team of 4, by discovering avenues of research and providing direction. Followed CRISP-DM principles. Conducted large-scale mining, parsing and analysis of information over a distributed network with 2 TB of data. Data Science Workbench migration - • Tasked with scripting high-fidelity and high-coverage field tests in Python to measure the speed, performance and bandwidth of the Cloudera data science workbench for an Hadoop ecosystem with 4 data nodes and 6 mining nodes.

Served as teaching assistant at the Department of Physics in UC San Diego, for the course Electricity and Magnetism lab - 1BL, and Waves and Optics - 1CL. Responsibilities included, conducting weekly lectures, proctoring and grading papers of 75 students.

This project aimed at implementing a full-scale Advanced Driver Assistance systems, including Pedestrian Detection, Lane Detection and Speed-bump detection. Effective systems which performs better on occluded Indian road conditions were sought after.Experience gained :• Implemented a HoG+SVM framework with SIFT models and used a back-propagating neural learning method for recursive adaptation.• Tested it against standard benchmarks, where our algorithm performed on-par against standard CalTech 101 dataset, but excelled under occluded Indian conditions. • Future work included implementing deep learning ConvNets and using random forests for better detection efficiency.

Published a journal paper : Srinath et.al, “Example-Based Super-Resolution”, Scientific World Journal, Article 8306342. Extracted similarity kernels using matrix valued operators and improved image quality by 2.1 dB and similarity by 22%.

As a winter intern, I worked on applications of DSP in sound equalization. I built an Adaptive Sound Equalizer using resonant frequency matching and predictive filtering using Kalman filters in Python and Embedded C in CCS Studio. I designed and developed an embedded system for music audio equalization using Blackfin BF609 processor. The equalizer achieved an 8.9:10 subjective equalization measure, 22% reduction in MSE.

• As a summer intern, I assisted a doctorate student in the research on formation of Bragg reflectors in tapered multi-mode graded index polymer optic fibers (GIPOFs). • I was responsible for chemical etching and tapering of fibers to form Bragg reflectors that can be used a medium to create small-scale optical based medical sensors.• Achieved a 0.9mm reduction in mean radius that improved the bandwidth of the sensor node around 21%

Master’S Degree, Electrical Engineering | Intelligent Systems, Robotics And Control

B.E, Electrical, Electronics And Communications Engineering

Aissce, High School, Mathematics And Computer Science