United States

Data engineering

AI/ML Engineering

Masters of Science

Bachelors of Engineering

Vivek Hariharan

ML Engineer

Communication

Problem-solving

Team player

Organisation skills

Leadership

Cloud computing

Machine learning

Python

Airflow

Apache Spark

Walmart Global Tech is the technology and business services organization within Walmart aiming to transform retail tech. As part of the Walmart Connects Retail Advertising Engineering team, I focused on building display advertising targeting techniques and internal data science platform tools.
- Led the data pipeline and recommendation model development for Macro Contextual Targeting expansion, an automated context expansion product feature designed to balance ad reach and conversion.
- Designed and socialized the data sketches plus calibration-based approach to estimate the size of rule-based targeting audiences for display advertising.
- Managed the internal A/B testing metrics pipeline and dashboards, aiding in both team-wide and executive business decisions.
- Designed and oversaw the development of an in-house data quality monitoring framework, enabling the team to identify issues sooner during on-call rotations.
- Managed workload of 3-4 data science engineers over two-week sprints, translated product requirements into machine learning designs, scoped solutions, and oversaw execution.
- Showcased best practices in scaling up Spark jobs and automating pipelines using Airflow
through newly initiated bi-weekly knowledge-sharing sessions.

Indeed, Inc. is an American worldwide employment platform for job listings. As part of the Job Seeker Org, I worked on generating models and insights exposed on the platform to help job seekers navigate the silence in the job search funnel after applying (internally coined as the Black Hole problem).
- Led end-to-end development of data/ML models that powered job seeker facing product features on Indeeds MyJobs page
- Developed a predictive model to determine the likelihood of a job application getting a response, used this model as the basis for inferring rejections at scale for job applications (~100K a week) to support companywide initiative of providing job seekers closure for job applications
- Developed data pipelines using Spark that generated insights about job applications (average employer response time/rate, employer sign of life). Exposing these as product features increased the repeat visitor rate of job seekers on the My Jobs web page (the place where job seekers manage their job applications) by 3%
- Established office hours and helped team members level up their analytical skills through 1:1 mentorship
- Trained engineers and fellow scientists on writing Spark jobs and working with large volumes of log data

Formation was a Dynamic offer platform company allowing marketers to create gamified offer experiences for their customers. I was an early employee primarily responsible for managing the data pipelines and personalization models for our then two big customers Starbucks and United.
- Early employee with a key role in scaling up the team and onboarding a major client (United Airlines).
- Developed an optimization pipeline using SoftMax exploration algorithm to match the right marketing offer to the right customer, maximizing net revenue for marketing campaigns.
- Created a proof of concept using contextual bandits (Vowpal Wabbit) for the optimization problem, to generate more individualized and targeted offers.
- Built the backend Spark data pipeline to manage data ingestion and manipulation for United Airlines.

[IPO’d then acquired by Adobe]
- Collaborated with the engineering team to make improvements to ad viewability and vCPM optimization strategies
- Generated weekly and daily executive reports on key metrics to assess state of ad viewability for the company

- Developed proof of concept using Vowpal Wabbit; showed increased accuracy in predicting outcomes (viewable view, click etc.) for an auction and collaborated to launch a new auction appraisal framework
- Designed and prototyped the in-house look-alike modeling pipeline to expand ad audience given a seed cookie list
- Collaborated with the senior scientist on using word embeddings (Word2Vec) to represent user behavior
- Automated feature extraction, model training and eval machine learning workflows for classifying cookies into demographic segments
- Optimized accuracy of segmentation models by iterating through linear learners, gradient boosting and feed-forward deep learning models
- Automated the provisioning of the transient cluster infrastructure using ec2 spot instances on AWS
- Researched and presented to the engineering org a study on targeting in a ‘cookie-less’ world

Vivek Hariharan🇺🇸