Jan 2022 – Jan 2024 · 2 yrs

• My role is to add ML to optimize and enhance the workflow across the products.
• I am also part of the Board to contribute to the vision of the company.

Jun 2020 – Mar 2023 · 2 yrs 9 mos

Mar 2020 – Jan 2022 · 1 yr 10 mos

• Leading Fairness Research for Mastercard AI Garage Team which resulted in the first Fairness research paper from Mastercard accepted in NIPS 2021 Data Centric AI Workshop
• Leading Bias Testing Effort to ensure all the models deployed and developed are fair and unbiased. Integrated Experian and D&B data to access protected attributes to bias test merchants and cardholder models. Also, worked with Credo AI startup to enable Fairness at scale.
• Built Account Intelligence scores to predict chargeback risk, gdv, fraud, crypto scores. Deployed the models in production by integrating it with real time card transactions. Lead the bias testing effort for these models.
• Developed a state-of-the-art hierarchical Bert-based model to enable transfer learning for transactional data by generating transaction-level features learned from sequential patterns concatenated with card-level features that can be used for any downstream tasks for best performance.

Nov 2018 – Mar 2020 · 1 yr 4 mos

• Idea was to build a root cause classifier using Telecom cell anomaly data converted into Image format. The initial model had only 91% F1 score. Since the model required anomaly data, it is hard to collect labelled anomaly data in Telecom. So came up with a GAN architecture to generate synthetic images such that generated images have new information for each anomaly class. SSIM measure is used to make sure generated images retains structural similarity from the real images. Came up with a 4 metric evaluation framework to evaluate GAN output. Using GAN generated images along with real images fetched 99% F1 score. Tried DCGAN, CGAN and WGAN variations. Filed an IP on this technology.
• SDN (Software Defined Networking) is heavily used in 5G to separate control and data plane. But SDN performance takes a toll when a networking event burst is punted to the controller and it might lead to DPN connect/disconnect and various other overheads. So came up with a ML assisted monitoring system, which checks current SDN state and if the health is diminishing then it fetches all the networking event info and start throttling these events to make sure SDN works efficiently. Filed an IP on this technology.

Jul 2017 – Nov 2018 · 1 yr 4 mos

• Developed NLP stack with an ensemble of classifiers (strong, moderate and weak) using Shallow NN, tf-idf matching for Intent Detection and Recommendation
• Developed Recommendation engine to respond back with a list of related possible intents in case of unmatched queries.
• Developed spell checker for banking domain and type ahead recommendations to tell User how to interact with AI engine (cognitive training).
• Developed unsupervised clustering algorithm to group unmatched queries to be analysed by bank rep to take data driven approach.
• Developed sentiment analysis algorithm using wordvec and 1D convolution to provide feedback.
• Contextual Analysis of the previous queries and intent responses to predict which bank product to upsell (advertisements and lead generation) along with a list of related intents.
• Developed Text Search Model to extract correct question answer pairs for partial input queries.

Jan 2016 – Jun 2017 · 1 yr 5 mos · Mumbai Area, India

• Leading software team for end to end implementation of the system.
• Developed Architecture of Miko (India's First Companion Robot) backend based on the feature requirements. Implemented end-to-end AI code.
• Developed ML-NLP code to understand User intent and respond according to it.
• Developed Recommendation Engine to assist user while interacting with Miko. It also generates suggestions for the not-used features so that user gets to know the full scope of Miko.
• Constantly improving algorithms by data mining and analysis of the User Input.
• Developed End-to-End encryption while interacting with Miko.
• Developed Entity Detection model using CRF for various new entities supported by Miko.

Jun 2013 – Jan 2016 · 2 yrs 7 mos · Bengaluru Area, India

• Worked on several MDM, MSM chipset verification environment and developed new test cases for design modules at core level verification and chip level verification.
• Developed verification environment from scratch for newly added design modules and identified and documented detailed use cases.
• Developed drivers for Qualcomm's Robotics Processor Dragonboard.

May 2012 – Jul 2012 · 2 mos · Bangalore

• Developed Gesture Recognition System from Seismic Sensors. Applied this System on various applications like Auto-Theft detection, Intrusion detection, Controlled Appliances of a Room
• Patent is filed for the Gesture Recognition System with Siemens

Jan 2012 – Jul 2012 · 6 mos

• Content Development for EnhanceEdu as a project under MHRD (Ministry of Human Resource Development)

May 2011 – Jun 2011 · 1 mo

• Satellite Image Processing to detect water bodies from a 7-band Satellite Image obtained from LANDSAT satellite and develop a MATLAB GUI for it.