Anand Jeyahar

Deployed Video analytics (voice and vision based NN models) for predicting (and visualizing) emotions during the course of an interview.

• Built and Demonstrated a couple of predictive models to predict upsells and account loss probability. • Designed, guided and helped setup end-to-end data warehouse (from distributed data sources), data pipeline, data-sanity checks, data dashboards and data modeling notebooks. • Built a randomforest model to predict customer attrition and reduced attrition rates, by designing pro-active customer engagement process.

• Reduced annotation time by half, by automating parts of the video annotation task • Reduced video segment loading time by the annotation tool, by replacing the technology used.

• Built a Retention analysis tool for checking and verifying weekly retention rates and helped decide features to build based on that • Improved the product(chat-app) quality(latency and content quality) by improving backend services, analytic tools, and some probabilistic algorithms for auto-moderation of images based on user flags. • Built a (tf-idf based for relevancy) ranking system for posts, based on recency, rarity and length(of text) in the post.

Reduced the time taken to switch router configs to one-tenth. Automated routing configuration changes in the network setup to load-balance(and re-route) during DDOS attacks or heavy loads for some specific domains,using a combination of django + fabric + triggers mechanism to issue the ssh commands.

• Reduced down-time in case of a server problems by setting up the database with replication for robustness, • Improved response time and reduced number of server-down incidents (even during peakload hours) by tuning mysql. • Improved response latency with a module manager for importing and executing python modules from the filesystem, querying db for pyc(python bytecode) files.

• Reduced the year-end balancing of fiscal batch time by customized the Finance Transaction session to allow upload of single balance transactions, currency-wise. • Baan V Customization:Report Layout modification for Sales Invoice layout of Hungary,china,chile. As a consequence of the RoHS Guidelines ECR,there were a few fields that needed to be printed based on whether the RoHS flag was true or false.

Artificial Intelligence, Computational Models of Cognition, Human Computer Interaction

Machine Vision, Image processing, Operations Research,