Research Areas:
- Natural Langauge Generation
- Chatbots and Dialogue Systems - Meaning-to-text Generation

▶2021-2022 MPhil in Advanced Computer Science, University of Cambridge, with Distinction
Thesis: User Simulator for Engaging Conversations on Wikipedia

▶2017-2021 B.S. in Computer Science, Carnegie Mellon University, University Honors
Minor in Mathematical Sciences

·Expanded features and functions for both SAQL (Salesforce Analytics Query Language)
and SQL used by customers and Tableau Salesforce Analytics Cloud to gather data for visualization.
·Implemented mathmatical functions end to end. The implementation ranged from parser,
to compiler, to optimizer, to computation. Added support for distributed queries.

▶August 2020-August 2018 Software Engineer Intern, Salesforce
·Trained, compared, and tuned several machine learning models that, given the user input dataset, can estimate the training time of different ML models offered on the Einstein Discovery platform, an AI powered data analytics tool.
·Designed and implemented the end-to-end framework, both frontend and backend, for runtime estimation using Python, Java, and Javascript.

▶June 2018 - August 2018, Software Engineer Intern, Mastercard
·Developed the multi language feature for PWR (Pay with Rewards) automation testing using Java, maven, and spring dependency injection. Reduced the testing span from 4 hours to under 45 minutes.
·Setup the Jenkins pipeline for multi device testing environment and improved the Jenkins configuration to maximize the testing flexibilities.

·Developed a reinforcement learning from human feedback framework to generate contentarticles for Cambridge Assessment English tests.
·Experimented with various prompt generation methods to better elicit implicitly encodedknowledge in large transformer-based language models.
·Develop an adaptive learning toolkit within the research area of Artificial Intelligence foEducation in collaboration with Cambridge Assessment.

▶2020.09-2021.03 Research Assistant, Carnegie Mellon University, DialRC Lab
·Trained different variations of the Recurrent Neural Network, such as seq2seq andtransformer, models to act as a user simulator in a dialog.
·Devised mechanisms for fine-tuned pre-trained open domain dialog systems, such asDialoGPT, to develop strong user simulators for goal-oriented dialogs.
·Use trained user simulator to generate realistic dialog data for the training of other dialog systems via reinforcement learning, and to act as an evaluation metric for these systems.

▶2019.06-2020.04 Research Assistant, Carnegie Mellon University
Simulating electron cry-tomograms using multi spherical model. And using simulatedtomograms to assess and optimize parameters for Difference of Gaussian particle pickingmethod.
·Participated in the development of AITom, an open-source AI driven cellular tomographyanalysis.

Given the public transportation schedule and historical trips in the Zurich area, this robust route planner will compute the fastest route between two stops within a provided uncertainty tolerance expressed as interquartiles. This project utilized Pandas for data manipulation, Markov Chain for the predicative modeling, and Kafka for streaming data.

▶2020 Machine Learning for Music Generation
Three different machine learning models (Hidden Markov Chain, Simple Recurrent Neural Network, and Performance Recurrent Neural Network) were trained with classical music pieces by well-known composers to generate new music pieces. To quantitatively evaluate the performance our model, we trained a convolutional neural network to classify the generated pieces into different composers.