About Me
Hello! I'm Aadith Sukumar, a passionate AI engineer and developer with expertise in creating robust deep learning models, debiasing AI systems, cyber security and integration.
With a background in Artificial Intelligence & Machine Learning along with a minors in CyberSecurity from Symbiosis Institute of Technology Pune, I bring a unique solutioning perspective to every project I work on. I'm dedicated to a result oriented thought process that solve real problems.
When I'm not developing, you can find me watching movies, playing badminton or acting. I believe in continuous learning and staying up-to-date with the latest technologies and trends in web development.
Experience
Working with the Platform Engineering team to develop and integrate various AI features in ITSM tools like ServiceNow. Integrating Generative AI and developing AI Agents for ServiceNow instances and enhancing it with automation solutions. Implementing CI/CD DevOps pipeline across platform instances to improve release frequency.
Part of a novel 'Multimodal Neurophysiological Framework for Cognitive Behavior Analysis' project funded by Govt. of India in collaboration with CDAC-Delhi and DRDO INMAS. Worked on building robust multimodal audio-video deep fusion models with 74% accuracy and collected primary sensor data. Leading a project on Adversarial Debiasing of Machine Learning models in Network Security for improving cyber-attacks prediction by 22.60%.
Simulated university network to build deployable network architecture and performed penetration testing for vulnerabilities. Proposed a detailed consultancy report with expert solutions for fixing 3 critical vulnerabilities and mitigating security risks by 65%.
Completed various projects including analyzing the web-security of web apps, performing VAPT, and implementing an encryption model for advanced steganography and information transfer.
Research Publications
Research project funded by Ministry of Electronics and Information Technology, Government of India in collaboration with CDAC Delhi on 'Multimodal Neurophysiological Framework for Cognitive Behavior Analysis'. A new multimodal dataset, CogniModal-D, is presented for deception detection, specifically targeting the Indian population and spanning seven modalities. Multimodal fusion approaches, integrating diverse verbal and nonverbal cues, significantly improve deception detection accuracy compared to unimodal methods.
This work focuses on Facial Emotion Recognition (FER) challenges due to biases in benchmarking datasets, model generalizability, especially in security and identification applications where disguises may deceive models trained on biased data. This study mitigates 'good-image' bias in datasets, proposing effective techniques like knowledge transfer and synthetic image generation, showing significant improvement in FER performance, validated on real-world data with a fourfold enhancement over baseline methods.
Funded collaboration with UAE-University, Abu Dhabi part of the United Nations SDG Research Grant. An innovative Artificial Intelligence of Things (AIoT)-based smart sensor approach to monitor the health status of the wind turbine blade. Multiple sensors are used to measure vibration, and cloud analytics based on machine/deep learning is utilized to make an informed decision on the turbine blade health/degradation.
Projects
Medium Articles
Contact Me
saadith2002@gmail.com
Location
Pune, India
aadith-sukumar