Udit Gupta

Leading exploration of Data Science innovations for applications related to pricing , sourcing, purchasing, financial forecasting, cashflow acceleration, and other related fields.

• Conceived, implemented, and deployed real-time point-processed based fraud-risk architecture. – Reduced the computational complexity of existing risk-factors from 𝑂(𝑛3 ) to 𝑂(1) – Reduced risk calculation cloud costs by 95%. • Implemented standard procedures for automated testing and measurement of deployed machine learning models – Released employee capacity from model monitoring and rectification. • Implemented algorithms for speaker identification, audio channel analysis, audio event detection and miscellaneous audio processing tasks for providing new fraud-risk engines and authentication engines for Pindrop’s products. • Interfaced with the engineering team for end-to-end collaboration for production implementation of forensic machine learning algorithms. • Successfully deployed Pindrop’s fraud-risk product in multiple customer environments to optimize the key performance indicators. • Communicated complex technical and scientific concepts to non-technical parties (customers) throughout the development process.

• Identified that existing state-of-the-art measures for blind source separation are not perceptually significant in the context of separating vocals and instrumental music from a musical mix. • Developed and proposed new prototype measures for evaluation of Singing Voice Separation. • Worked towards improving the proposed prototype measures and demonstrating statistical significance and robustness of improved measures. • Developed systems and tools to be used by child psychologists for early detection of autism or developmental defects in children. • Developed algorithms for detection and diarization of speech, tantrum detection, and detecting and analyzing child directed speech from day-long audio-recordings. • Collaborated across medical and engineering researchers from four research institutions.

• Developed a prototype for real-time and predictive customer sentiment analysis for call center based telephonic customer interactions to predict the likelihood of churn. • Prototype was used by the research team for client specific customization and deployment.

• Implemented cost effective Ambisonic encoding and decoding system for an auditorium. • Engineered a software tool to place monophonic audio recordings in any virtual environment and generate Ambisonic encoded spatial audio and a play it back to custom speaker configurations.

• Collaboratively developed an image similarity based query system for the fashion industry. • Researched and modified dimensionality reduction algorithms using sparse methods to achieve a balance between accuracy and computational complexity. • The adapted algorithm reduced the processing time by thirty to forty percent with negligible loss in accuracy.

• Investigated methods of enhancing communication bandwidth of existing optical fiber infrastructure of Korea. • Researched and proposed the use of digital modulation techniques commonly used and available for wireless communications. • Presented a report on the requirements for optical transmitters and receivers to implement these techniques on an optical network.

Trainee in the Optical Fiber Comission Link Setup Project.