Senthil Kumar Kumarasamy Mani Email & Phone Number

Bengaluru, Ka, In

Sunnyvale, Ca, Us

[2021 - 2024] We work in partnership with AI and sister teams in AI infrastructure and facilitate robust, efficient, straightforward application of machine learning capabilities to LinkedIn’s mission. We are one of the core teams behind the Productive Machine Learning (Pro-ML) initiative, bringing a fast end-to-end model and feature goodness to LinkedIn engineering. Our major projects include:Health Assurance: Aims to provide a lightweight, automated, and efficient way to validate & monitor models at inference time as well as catch and raise the visibility of problems as early as possible. It is a key pillar of Productive Machine Learning (Pro-ML)Model Deployment: The process of allowing a trained model to be ready for use in production. It represents the entire timeline from when a model is finished training until the application code can use the model. Pro-ML model deployment was designed and built to provide a common process and set of tools to allow Pro-ML engineers to easily deploy their models and monitor the health of each model. Pro-ML model deployment provides a suite of tools that enable users to create, track and deploy (or un-deploy) model artifacts to production.Model Cloud: Serving models for point/batch inference is inarguably operations-heavy and requires teams to build/integrate with the ProML’s serving ecosystem. While this adds complexity at the application layer, it requires non-trivial engineering investment for each team and consequently prevents creating leverage. To address this problem, we are building a managed model-serving solution (Model Cloud) for online inference

Bangalore, Karnataka, In

Developing a platform for managing the complete lifecycle of AI Model development across data sourcing / management, pre-processing, augmentation, training / experimentation, testing, hosting and monitoring, with data science intelligence baked into the platform.Developing an Enterprise Knowledge Graph Platform, to enable storing and serving of semantic information about product and addressing selection gap through advanced search and recommendationDeveloping an embedding based retrieval platform to store and index multiple embeddings of the product and power solutions like duplication detection, visually similar products and shop the look.

[ 2019 ] - Application Modernization. Developing AI-Based tools and solutions to help modernize monolith applications to be cloud-native. Specifically, focus on recommending micro-services form monolith applications and a modernization assistant which can help in migration of third party dependencies. We are using a combination of natural language processing, machine learning, and program analysis. As part of this work, we have filed 2 patents. [ 2017 - 2019 ] SE4AI - Software Engineering for AI. There is no structured / principle way of building AI applications. Current SE principles and methodologies have several gaps, in all aspects of requirements, design, implementation, test and deployment when it comes to applying it to building AI applications. We have started investigating this space towards defining process and methodologies and associated tools required to improve the productivity of AI developers, to truly democratize AI. Toward this goal, we have reduced the barrier to entry for developers by building DARVIZ (internal project name), which allows a developer to visually program a DL model and generate the model code in any of the libraries (TF, Keras, Caffe and Pytorch). This work is now offered as Neural Network Modeller (NNM) in Watson Studio product. As part of this work we have filed 3 patents and 2 publications (ICSE'16 and AAAI 18)There is a need for both AI developers and Business SMEs to thoroughly test the AI models prior to deploying them in production. Testing AI models is very nascent, with limited process and tooling support and is highly subjective to the skill level of testers. We developed an AI Testing Framework that automates AI Testing by generating and executing test suites that meaningfully represents data in the wild. We have developed techniques for testing the sensitivity and adversarial robustness of AI Models. We filed 2 patents and published 1 paper (InterSpeech 2019, ICSE 2020 & AAAI 2020 - submitted)

[2015 - 2017] AI Based Automation - The goal of AI based Automation is procedural assistance either as a self–assist solution (assisting the end-users directly) or as agent-assist solution, with the value proposition of increasing productivity and cost savings with efficient problem resolution. In building such a system there exist considerable technical challenges. The knowledge required for training any cognitive system, is not available in well-documented high fidelity format. Practitioners over the years have gained the expertise of resolving issues and this tacit knowledge is not documented in any format. Further, there exist a huge variation across the users in reporting the same issue. The issues can be reported in a very ambiguous ways and requires disambiguation through manual intervention. Thus a AI based general assistance solution needs to be able to handle such linguistic variations, disambiguate the context, and learn and self adapt through its usage. We built a AI platform IBM Automation with Watson hosting automation specific services, and solutions like Employee / User Assist and Agent Assist on top of this platform. Filed 3 patents and published 3 papers ( IBM R&D Journal'17, ESEC/FSE'17 and AAAI'18)

[2012 - 2014] Catapult Preventive : Applied machine learning and analytics for deriving insights from mining software repositories like ticketing systems, and code version history - to help stakeholders identify preventive actions to reduce ticket volumes and do effective transitioning in application maintenance space.Filed 4 Patents and Co-authored 12 papers in peer reviewed conferences

Investigated the requirements management method used in packaged applications (SAP and Oracle) and developed solutions to improve asset reuse and management. Primarily responsible for-- Design and Implementation of innovative solutions-- Led and Mentored a team of developers-- Successfully transfered the research solution to business groupsFiled 4 Patents and Co-authored 11 Papers in peer reviewed conferences

Worked on realization of Model Driven Approach for Software Development. Developed Eclipse framework based tools for process and UI modeling and automatic realization of the same applying MDD principles. Primarily responsible for- Designing Innovative Solutions through active research and implement them- Successfully transfer research solutions to business groupsCo-authored 4 papers in peer reviewed conferences

Bangalore, Karnataka, In

Bangalore, Karnataka, In

B.E & M.Sc, Computer Science & Mathematics

Education record