Vinod Sharma Email & Phone Number

Gurugram, Haryana, India

Launched Trust & Safety AI platform to streamline and scale human content moderation workflows at reduced cost leveraging AI Copilot

Gurgaon, Haryana, India

Grew the data science team from 0 to 6 to implement AI/ML strategy for NRS group with Annual revenue of $400M+Oversaw a team responsible for developing 30+ machine learning models, enhancing advanced analytics capabilitiesReduced operational cost by 60%($3.6M annually) by delivering a scalable Document AI platform to extract insights from documentsScaled Document AI platform to 15M Pages annually by implementing microservices in AWS Cloud(APIGateway, Lambda, SQS… Show more Grew the data science team from 0 to 6 to implement AI/ML strategy for NRS group with Annual revenue of $400M+Oversaw a team responsible for developing 30+ machine learning models, enhancing advanced analytics capabilitiesReduced operational cost by 60%($3.6M annually) by delivering a scalable Document AI platform to extract insights from documentsScaled Document AI platform to 15M Pages annually by implementing microservices in AWS Cloud(APIGateway, Lambda, SQS, SageMaker)Optimized document insights generation time from 1 hour to 10 minutes by delivering 20+ models with 90+% accuracy leveraging Deep neural networks(LayoutLM, Azure Forms & BERT) and Implementing Retrieval Augmented Generation(RAG) with a Vector database and Azure OpenAIDesigned and developed an NLP-based multi-class classification model with 92% accuracy using pre-trained BERT to map questions to control objectives reducing the time from 2 hours to 10 minutesDesigned and trained a handwritten page classification model with 99% accuracy using a pre-trained Xception model Show less

Noida Area, India

Led a team of 4 to build models and dashboard/reports, enhancing analytics capabilitiesDesigned and developed a Seat Availability prediction model using XGBoost, reducing SMAPE by 20%, serving 200K daily requestsTrained and deployed a binary image classifi cation model with 91% accuracy using pre-trained Xception model to detect bad food packaging

Palo Alto, California

Increased orders by 2% by designing and developing a personalized meal recommendation engine using hybrid (content and collaborative based) filtering method Increased user retention by 3% by creating a churn prediction model using logistic regression with 89% accuracy. Improved the accuracy to 93% by using random forest Reduced time to build continuous integration and continuous deployment pipeline for data science applications from 4 weeks to 1 hour by architecting and… Show more Increased orders by 2% by designing and developing a personalized meal recommendation engine using hybrid (content and collaborative based) filtering method Increased user retention by 3% by creating a churn prediction model using logistic regression with 89% accuracy. Improved the accuracy to 93% by using random forest Reduced time to build continuous integration and continuous deployment pipeline for data science applications from 4 weeks to 1 hour by architecting and developing a general framework Show less

Greater Los Angeles Area

Designed a user tracking mechanism in Bluetooth domain by formulating a fingerprinting problem as a clustering problem. Implemented using Python, NumPy, SciPy, Scikit-learn.Developed a data collection service using Python/Django, Postgres, Heroku, AWS, iOS(Swift), Android(Java) and Git. Designed a privacy-preserving data mining framework using distributed k-anonymity. Prototype implemented using Python, SOCKS, Apache, MySQL and SQLite. Designed a defense… Show more Designed a user tracking mechanism in Bluetooth domain by formulating a fingerprinting problem as a clustering problem. Implemented using Python, NumPy, SciPy, Scikit-learn.Developed a data collection service using Python/Django, Postgres, Heroku, AWS, iOS(Swift), Android(Java) and Git. Designed a privacy-preserving data mining framework using distributed k-anonymity. Prototype implemented using Python, SOCKS, Apache, MySQL and SQLite. Designed a defense mechanism for DDoS attack by solving an anomaly detection problem to distinguish human users from malicious bots. Developed using Python and Scikit-learn.Teaching Assistant: Foundations of Artificial Intelligence, Database Systems, Operating Systems, Introduction to Computer and Network Security. Show less

San Francisco Bay Area

Designed and developed a validation framework for a safe web browsing scheme using distributed crawler. Implemented using Python, PhantomJS and RabbitMQ.

Bangalore, India

- Worked in the Application Firewall Research Lab.- Designed and developed next-generation application firewall gateway.- Researched 100+ application protocols used for network communication to identify application signatures.

Noida Area, India

- Worked as part of Vulnerability research team of Security Research Group.- Researched vulnerabilities to identify possible intrusion detection and prevention signatures.- Developed proof of concept and writing advisories on newly published vulnerabilities.

Master'S Degree, Computer Science, 3.63/4.0

B.Tech, Information Technology