Close

Anup Shakya

Research Scientist | Meta

Download Resume

About Me

Namaste!!! 🙏 Welcome to my portfolio!!


I'm Anup Shakya, currently working as a Research Scientist at Meta, where I design and build ranking and recommendation systems that scale to very large populations and are tuned to business and product requirements. I have also worked on building evaluation for multi-turn speech models, including post-training and fine-tuning models. I have a PhD in Computer Science from the University of Memphis. My passion lies in crafting solutions to enhance the field of Neuro-symbolic learning. My current research interests are in Machine Learning, Recommendation Systems, and Data Mining. I am also interested in Natural Language Processing, Computer Vision, and Generative AI. I also have a Master's Degree in Computer Science.

As a seasoned software engineer, I bring a wealth of experience in Java, Python, C++, debugging, critical thinking and problem-solving with a robust understanding of HIPAA compliance in the U.S. health sector.

I just love MUSIC and don't really have a preference for a music genre. I like to play the guitar and sing. I also love to run and hike. I am a big-time chai ☕ lover.

Please feel free to explore my portfolio to discover my academic and professional journey.

Work Experience

Meta

Research Scientist

  • Design and build ranking and recommendation systems that scale to very large user populations and are tuned to business and product requirements
  • Build evaluation for multi-turn speech models, including post-training and fine-tuning of models

Rivian Automotive

Software Engineering Intern

  • Researched tools and techniques to optimize large language models such as LLaMa and OpenAI Whisper for deployment in edge devices
  • Worked to enhance the throughput of Rivian voice assistant framework by testing the optimization strategies like quantization, knowledge distillation and utilizing Deep Learning Accelerators (DLA) on NVIDIA Jetson AGX Orin

University of Memphis

Graduate Research Assistant

  • Gained real-world research experience working on several research projects funded by National Science Foundation
  • Developed the ability to effectively contribute in a collaborative team-oriented environment leading to 3 successful research collaborations
  • Secured 2 fundings including one from the Bill & Melinda Gates Foundation
  • Exhibited ability to investigate, evaluate, and progress solutions with an ability to write scientific articles as evidenced by 6 peer-reviewed publications

University of Memphis

Graduate Teaching Assistant

  • Organized the class hours and discussion sessions with students.
  • Graded the assignments, class works, projects, and final examination paper.

Deerwalk Services

Software Engineer (Team Lead)

  • Highlighted ability to maintain high coding standards with expertise in building web applications leading to the development of 4 in-house web application products
  • Led software development and enhanced the agile development process resulting in a 25% improvement in efficiency evidenced by lead development role in 2 web application projects

Deerwalk Services

Associate Software Engineer

  • Practiced high-quality coding standards following specific design patterns and developed / maintained web applications.
  • Learned to debug issues in a large-scale web application and meet product requirements.

Education

University of Memphis

Jan 2020 - Dec 2024

Ph.D. in Computer Science

Graduation Date: Dec 15, 2024

University of Memphis

Jan 2020 - May 2022

MS in Computer Science

Institute of Engineering, Central Campus Pulchowk

Nov 2011 - Aug 2015

Bachelor's in Electronics and Communication Engineering

Aggregate grade: 74.47%

Research Publications

Anup Shakya, Abisha Thapa Magar, Somdeb Sarkhel and Deepak Venugopal, Reparameterizing Hybrid Markov Logic Networks to handle Covariate-Shift in Representations, In Proceedings of 41st Conference on Uncertainty in Artificial Intelligence (UAI '25), 2025 July. Link to Paper

Abisha Thapa Magar, Anup Shakya, Somdeb Sarkhel and Deepak Venugopal, " Can A Language Model Represent Math Strategies?": Learning Math Strategies from Big Data using BERT, In Proceedings of 15th International Learning Analytics and Knowledge Conference (LAK '25), 2025 Mar, pp. 655-666, doi:10.1145/3706468.3706558. Link to Paper

Anup Shakya, Abisha Thapa Magar, Somdeb Sarkhel and Deepak Venugopal, On the Verification of Embeddings with Hybrid Markov Logic, In Proceedings of IEEE International Conference in Data Mining (ICDM), 2023 Dec, pp. 1301-1306, doi:10.1109/ICDM58522.2023.00165. Link to Paper

Abisha Thapa Magar, Anup Shakya, Somdeb Sarkhel and Deepak Venugopal, Verifying Relational Explanations: A Probabilistic Approach, In Proceedings of 2023 IEEE International Conference on Big Data, Sorrento, Italy, 2023 Dec, pp. 108-115, doi:10.1109/BigData59044.2023.10386213. Link to Paper

Anup Shakya, Vasile Rus and Deepak Venugopal, Scalable and Equitable Math Problem Solving Strategy Prediction in Big Educational Data, In Proceedings of 16th International Conference on Educational Data Mining (EDM), International Educational Data Mining Society, 2023, pp. 137-148, doi:10.5281/zenodo.8115669. Link to Paper

Anup Shakya, Vasile Rus and Deepak Venugopal, Mastery Guided Non-parametric Clustering to Scale-up Strategy Prediction, In Proceedings of 37th AAAI Conference on Artificial Intelligence, Workshop on Artificial Intelligence for Education, 2023 Feb. Link to Paper

Anup Shakya, Vasile Rus and Deepak Venugopal, Student Strategy Prediction using a Neuro-Symbolic approach, In Proceedings of 14th International Conference on Educational Data Mining (EDM), International Educational Data Mining Society, 2021, pp. 118-129. Link to Paper

Anup Shakya, Vasile Rus, Stephen Fancsali, Steve Ritter and Deepak Venugopal, NeTra: A Neuro-Symbolic System to Discover Strategies in Math Learning, In Proceedings of The Third Workshop of the Learner Data Institute, The 15th International Conference on Educational Data Mining (EDM 2022), 2022. Link to Paper

Deepak Venugopal, Vasile Rus and Anup Shakya, Neuro-Symbolic Models: A Scalable, Explainable Framework for Strategy Discovery from Big Edu-Data, In Proceedings of the 2nd Learner Data Institute Workshop in Conjunction with The 14th International Educational Data Mining Conference, 2021 Jun. Link to Paper

Anup Shakya, Towards the robustness of Transformer Models on Image Data, http://dx.doi.org/10.13140/RG.2.2.23552.10242, 2022. Link to Paper

Projects

Probabilistic Verification of Neural Networks | (Link to Project)

Relevant Skills: Python, PyTorch, scikit-learn, Hybrid Markov Logic Networks, MILP Optimization, Piecewise-Linear modeling, Hypothesis Testing

Proposed a novel approach to verify representations in Deep Neural Networks beyond standard classification tasks. Introduced a framework based on Hybrid Markov Logic Networks (HMLNs), allowing for the specification of complex properties combined with symbolic domain knowledge. Developed a method to learn parameters and a verification process using Mixed Integer Linear Programming, showcasing the versatility of the approach across applications like Graph Neural Networks, Deep Knowledge Tracing, and Intelligent Tutoring Systems.

Scalable Student Strategy Prediction in Math Learning | (Link to Project)

Relevant Skills: Python, TensorFlow, Transformers, LSTM, Pandas, Learning Science, ML Optimization

Developed an innovative embedding, MVec, and employed non-parametric clustering to achieve scalable and fair strategy prediction. Demonstrated exceptional accuracy and predictive equality in large-scale student interaction datasets from MATHia.

Towards the robustness of Vision Transformers | (Link to Project)

Relevant Skills: Python, PyTorch, Vision Transformers, Generative Adversarial Networks, Data Augmentation, Computer Vision, Classification

Drove innovation in computer vision with "Vision Transformer on Image Data." Conducted a rigorous study on the model's robustness to noise, showcasing its potential to outperform convolution-based models. Implemented simple data augmentation techniques, unveiling their significant impact on enhancing generalization and overall performance.

Learning to play flappy bird using DDQN | (Link to Project)

Relevant Skills: Python, Reinforcement Learning, DQN, Double DQN, simulation

This project explores the realm of reinforcement learning, leveraging the potent Deep Q-Networks (DQN) algorithm to enable an agent to master the challenging Flappy Bird game. Despite facing high-dimensional sensory input and no prior knowledge of the game's elements, the DQN algorithm excelled in learning optimal strategies, ultimately achieving super-human performance. The project delves into the intricacies of learned representations, addresses challenges, and offers insights into the capabilities and potential enhancements of DQN in the context of gaming.

Skills

Get in Touch