Gowtham Ganesan Email & Phone Number

Hyderabad, In

Bengaluru, Karnataka, India

Enhancing Test Coverage with Generative AI: A Multi-Stage LLM Approach for Automated Regression Testing• Replaced conventional Search Based Software Testing (SBST) methods with Generative AI to generate test cases using a multi-stage prompt strategy, resulting in better test coverage. Achieved 4× improvement in pass rate, along with a 10% increase in line coverage and a 4% increase in branch coverage compared to the conventional method.• Leveraged the Meta Llama 3.1 Large Language Model with the LangChain framework to automate regression test creation and enhance fault coverage.• Improving the LLM's performance by exploring fine-tuning with PEFT (LoRA, Soft prompt), and reinforcement learning with human feedback techniques.

Bengaluru, Karnataka, India

Deep Learning-Based Regression Test Prioritization for CI Testing• Designed a deep learning model to prioritize test cases, considering both time efficiency and fault detection effectiveness. The test cases encompass all stages of RTL synthesis, including power optimization, satisfying area and timing constraints etc. This led to an 81% improvement in the NAPFD (Normalized Average Percentage of Faults Detected) score compared to the baseline model.• AWS SageMaker was utilized to build the ML model along with Amazon S3.• Explored data augmentation methods, including SMOTER and SMOTER-GN, to manage data imbalance.

Hyderabad, Telangana, India

Leveraging Graph Neural Networks for Identifying Security Vulnerable Code Segments• Constructing an ML model for identifying vulnerable APIs achieved a 73% recall score. In Qualcomm, non-ML approaches reduced cyber security attacks by 37%. Combining ML with non-ML methods led to an extra 15% reduction, totaling a 52% decrease in attacks. • Improvised the performance of the ML model by exploring Graph LSTM and Graph Convolution for the Graph Neural Networks.• Utilized pre-trained programming language models - CodeBERT and GraphCodeBERT, as the embedding layer, resulted in faster model convergence, thereby saving time and reducing the need for large dataset for training the ML model.Harnessing Deep Learning for Hierarchical Document Classification• Employed Deep Learning techniques to categorize cryptographic documents within Qualcomm, facilitating rapid retrieval of relevant documents based on user queries. This streamlined the onboarding process for new team members and allowed experienced professionals to gather information efficiently.• Achieved 20% higher accuracy with Deep NLP models compared to conventional ML methods.

India

• Led the development of a Python framework from inception to completion, encompassing requirements gathering, design, prototyping, and implementation. This framework simplifies automated feature testing for upcoming builds and was showcased at a Qualcomm Tech talk.• For each chipset, this tool contributed to a 24-week reduction in L0 validation effort for developers.

Coimbatore Area, India

It was an internship on Android. All the essential concepts and tools to develop an Android application were explained. At the end of the internship, along with my team mate, I developed a signature based entry recording app which could be used during examinations to record the students who have appeared for the examination. This app's purpose was to reduce the usage of papers and use a mobile app in place of it. So, the cost incurred for the examinations could be saved. SQLite Database has been used to store and retrieve students' information.

Coimbatore Area, India

In this internship, Java, HTML and JavaScript programming languages were taught. At the end of this internship, Online Quiz Management System application was developed. In this application, multiple choice questions will be presented to the user and evaluation of the options chosen by the user is done by comparing with the correct options stored in server.

Master Of Technology - Mtech, Artificial Intelligence, 9.62 / 10.0

IIIT-Delhi has provided me with a comprehensive AI education, blending theory with practicality. Mandatory machine learning projects in AI.

Bachelor'S Degree, Computer Engineering, 9.75 / 10.0

Activities and Societies: National Aeronautics and Space Administration(NASA) conducted a special training program in Amrita Vishwa.

Education record