Jul 2021 – Apr 2025 · 3 yrs 9 mos · Palo Alto, California, United States

• I worked on the aircraft robot motion planning capabilities to improve flight systems and automation.

Jan 2021 – Jul 2021 · 6 mos · Pittsburgh, Pennsylvania, United States

• Researched robotics, motion planning and machine learning based methods to evaluate and improve the motion planner's performance.
• I also worked on improving the performance of the motion planner in cases where the ego vehicle follows a lead vehicle.

Mar 2019 – Jan 2021 · 1 yr 10 mos · Pittsburgh, PA

• I worked on designing and implementing robotics and machine learning based algorithms to predict the behavior of actors in the environment around the self-driving vehicle. I also led several mission critical projects for the development of dynamically feasible trajectories through a structured prediction problem formulation, some of which included collaboration with the Motion Planning team.
• Some examples of my work (published only) include
• Improving Movement Predictions of Traffic Actors in Bird’s-Eye View Models using GANs and Differentiable Trajectory Rasterization (https://arxiv.org/pdf/2004.06247.pdf)
• Long-term Prediction of Vehicle Behavior using Short-term Uncertainty-aware Trajectories and High-definition Maps (https://arxiv.org/pdf/2003.06143.pdf)
• Ellipse Loss for Scene-Compliant Motion Prediction (https://arxiv.org/abs/2011.03139)
• Physically Feasible Vehicle Trajectory Prediction (https://ml4ad.github.io/files/papers2020/Physically%20Feasible%20Vehicle%20Trajectory%20Prediction.pdf)
• Investigated the application of inverse reinforcement learning for the task of vehicle motion prediction

May 2018 – Aug 2018 · 3 mos · Pittsburgh

• As a software engineering intern in the Prediction team, I worked on improvements to the deep learning based prediction models both in terms of performance accuracies and latency that have an overall impact on the entire autonomy SW stack
• Trained, analyzed and landed a new deep learning model with improved prediction accuracies (all of several metrics) and 30% lower 99th percentile latency
• Designed, implemented (across the entire prediction SW stack), trained and analyzed a single deep learning model that unifies several existing models to achieve prediction accuracies similar to production, a 10x reduction in overall parameter count and 18% lower median latency
• Designed, implemented and landed key metrics to evaluate performance of production deep learning based prediction models

Sep 2015 – Jun 2017 · 1 yr 9 mos · Suwon, Gyeonggi-do, Korea

• Research, development and prototyping of Machine Learning/Artificial Intelligence/Deep Learning driven Big Data solutions for the Networks Business - unsupervised anomaly detection and time series prediction (TensorFlow) for 4G Key Performance Indicator data, root cause analysis (>95% accurate), unsupervised indoor-outdoor detection of user elements, supervised location estimation of user elements from radio frequency information (median ~45m), outlier removal and model fitting for capacity design on massive datasets generated by the Samsung LTE network equipment logs
• Designed and developed an efficient algorithm (log-linear time complexity) for detecting handovers for calls for Samsung’s VoLTE Monitoring and Analysis (VoMA) tool, deployed Pan-India for Reliance Jio
• Developed Python based network data analyser tool to monitor and detect issues from VoMA data, deployed Pan-India for Reliance Jio
• Managed and led a Software Centre, Seoul team for the research and development of a boosting model based root cause analysis solution for VoMA on LTE data from Mumbai areas
• Worked closely with a team of engineers and product designers to create the core features of Samsung CognitiV Analytics (proprietary Mobile Big Data Network Analytics and Optimization Platform); designed the algorithms and features such as KPI Analysis and guided the development team for successful implementation of the same
• Linux Administration - automated periodic transfer of data from LTE elements in India to Samsung HQ, Korea; analysed peak memory usage times of data collection tool and optimised tool start-schedule

May 2014 – Jul 2014 · 2 mos · Bangalore

• Designed and developed statistical solutions for portfolio management as a summer analyst in the market risk segment of divisional strats
• Implemented a risk management heuristic to aid the firm’s risk managers’ decisions based on VaR, PCA, MPC
• Set up infrastructure for exposure data (PNL, deltas) of the firm’s portfolios, enabling graphical analysis of products under various asset classes; identified correlation between key market variables, firmwide portfolios

Dec 2013 – Dec 2013 · 0 mo · Hyderabad Area, India

• Studied and implemented a momentum strategy (Jagadeesh and Titman’s 1993 paper) on optimal portfolio construction; processed historical data to study returns of the winner-minus-loser portfolio over different time horizons
• Studied, implemented the K-means clustering algorithm, SVMs for credit scoring