I work on production-grade Machine Learning compiler technologies, with a focus on Neural Engine accelerator compilation, performance modeling, CI, and tooling. My work includes contributing to TOSA MLIR, reference models, and compiler toolchains, while collaborating across teams to enhance support for the Arm architecture and broader MLIR ecosystem.

1. Enhanced IC design processes by developing and maintaining software validation tools for complex Process Design Kits (PDKs) in Cadence Virtuoso Studio using C++ and Cadence SKILL language; ensured high-quality PDK validation and migration through robust, integrated solutions within the design environment.

2. Improved PDK validation flows by designing, implementing, and testing integrated software components; Formulated design, functional, and test specifications; developed essential features for robust, repeatable, and reliable PDK qualification; improved performance, quality, and stability of PDK validation software through the integrated regression test systems for automated unit testing, enhancing code quality and facilitating development.

3. Enhanced tool performance and customer experience by developing user-friendly UI features, streamlining batch operations with foundational APIs, implementing robust testing frameworks and regression systems, and resolving critical customer-reported issues—resulting in reduced processing errors, and improved software stability. For backend, improved database structure, and internal data synchronisation, leading to increased efficiency, performance, and maintainability.

4. Collaborated with customers to implement product enhancements and resolve development bugs, aligning solutions with customer needs; improved software quality and maintainability by addressing challenges in PDK migration and analysing PDK quality and readiness for other Cadence products.

1. Implemented Diffusion-Based Text-to-3D Models: Deployed and optimized Stable Dreamfusion, a state-of-the-art diffusion-based neural network model, to generate accurate, high-fidelity, and textured 3D objects from text and images, achieving and enhancing 3D modelling capabilities with data-driven solutions.

2. Integrated Advanced 3D Modeling Techniques: Studied and combined strengths from GANs, ControlNet, Pixel2Mesh and OpenAI's Point-E models to improve 3D object generation; identified effective components from each to enhance model performance and output quality.

3. Optimised model performance through extensive experimentation with different neural network architectures, conducting hyper-parameter tuning to improve the quality and accuracy of generated 3D models.

4. Implemented custom evaluation metrics (Chamfer Distance and Earth Mover's Distance) to assess reconstruction quality, ensuring high-fidelity and human-acceptable 3D point clouds.

5. Deployed cloud-based training pipelines on AWS SageMaker, utilizing Docker to streamline and automate the training process on cloud infrastructure. Gained experience in deploying machine learning models using AWS SageMaker, MLFlow, and Terraform, and assisted in setting up APIs with AWS Lambda for seamless integration.

6. Collaborated closely with the Data Science team to integrate models into existing pipelines, define project requirements, and communicate progress to my mentor and manager.

7. Analysed the capabilities of 3D object generation models to uncover new applications within product lines, providing insights for future development roadmap.

I worked as a Mentor (Teaching Assistant) for undergraduate course 'System Design Project'. I was assigned mentor to two groups to guide their progress in designing assistive robots as prototypes for potential industry application. As a mentor I am responsible for helping your group keep on track, to self-organise and assist with organisation related questions. I had separate group meetings with my two groups to facilitate their project management and progress and come up with next steps together, sometimes groups consulted me about technical aspects of their project. Other responsibilities include being responsive and resonable to additional enquiries from my groups and keep track of individual students’ participation in the group and give formative feedback if contribution seems uneven. I had given comprehensive and inclusive feedbacks to their intermediate and final reports in great details. The feedback, expertise and guidance inspired and helped my groups for good final project delivery.

IPAB Research Internship: Hierarchical control based on differential extrinsic plasticity

This summer research internship was supported by the Institute of Perception, Action and Behaviour (IPAB) of University of Edinburgh and was funded at the UE3 grade.

We propose a hierarchical control scheme that is based on a neural synaptic rule (DIAMOND-DEP) which is useful in generating periodic behaviour. Our novel mechanism is biologically plausible in nature and can lead to a new understanding of the emergence and convergence of the periodic behaviours in evolution. Evolutionary autonomous development presented by periodic behaviours is achieved in our work by a brain equipped with deep neural network.

• Proposed a hierarchical control scheme that is based on a neural synaptic rule useful in generating periodic behaviour.

• Proved that our novel mechanism (DIAMOND-DEP) is biologically plausible in nature and can lead to a new

understanding of the emergence and convergence of the periodic behaviours in evolution.

• Achieved evolutionary autonomous development presented by periodic behaviours by a brain equipped with DNN.