Aadith
Sukumar
Building secure, scalable AI systems, integrating generative AI frameworks in enterprise systems at Maersk. Researcher in multimodal ML, adversarial debiasing, and AI security — with papers in Nature and IEEE.
About Me
Aadith Sukumar
Associate Software Engineer · A.P. Moller - Maersk
AI/ML Engineer and researcher based in Pune, India — specializing in Agentic AI, multimodal machine learning, enterprise tools and cybersecurity.
I work on enterprise IT landscape tools, building secure, scalable cross-platform integrations and AI systems at A.P. Moller - Maersk, developing Generative and Agentic AI into ServiceNow workflows and orchestrating DevOps pipelines for global deployments — improving release frequency by 250%.
My research at SCAAI spans multimodal deception detection (Govt. of India MeitY grant), adversarial debiasing in network security, and facial emotion recognition. Published in Nature Scientific Reports and presented at multiple prestigious international conferences.
Outside engineering, I led the AI Club and Cybersecurity Clubs at SIT Pune, mentor students and am a theatre & film enthusiast.
Skills & Tech Stack
AI & ML
Engineering
Languages & Tools
Security
Experience
Associate Software Engineer
CurrentA.P. Moller - Maersk
Key member of the ServicePlatform team, specializing in Agentic AI, ServiceNow automation, and enterprise integrations. Engineered robust internal APIs and orchestrated DevOps pipelines that drove a 250% increase in release frequency and significantly reduced manual IT overhead globally.
Engineering Intern
A.P. Moller - Maersk
Engineered foundational AI workflows in ServiceNow and built a novel CI/CD pipeline across multiple instances while integrating 5+ Generative AI frameworks, bridging GenAI capabilities with existing enterprise infrastructure.
Research Intern — MeitY Grant
SCAAI · Symbiosis Centre for Applied Artificial Intelligence
Part of 'Multimodal Neurophysiological Framework for Cognitive Behavior Analysis' funded by Govt. of India (CDAC-Delhi & DRDO INMAS). Built multimodal audio-video deep fusion models achieving 74% accuracy. Presented FER research at FG2024, Istanbul. Led Adversarial Debiasing project improving cyber-attack prediction by 22.60%.
Project Intern
Cisco
Simulated a full university network architecture and conducted penetration testing to uncover vulnerabilities. Proposed a detailed consultancy report with expert solutions, fixing 3 critical vulnerabilities and mitigating security risks by 65%.
Cyber Security Intern
Academor
Analyzed web-security of applications, performed VAPT, and implemented an encryption model for advanced steganography and secure information transfer.
Publications
Peer-reviewed work in Nature, IEEE, and international conferences across AI, cybersecurity, and healthcare.
Multimodal Machine Learning for Deception Detection using Behavioral and Physiological Data
Research funded by MeitY, Govt. of India (CDAC-Delhi & DRDO INMAS). Introduced CogniModal-D, a new multimodal dataset for deception detection spanning 7 modalities targeting the Indian population. Demonstrated that multimodal fusion approaches significantly outperform unimodal methods.
Training Against Disguises: Addressing and Mitigating Bias in Facial Emotion Recognition with Synthetic Data
Mitigated 'good-image' bias in FER datasets using synthetic image generation and knowledge transfer. Achieved fourfold enhancement over baseline methods, validated on real-world security scenarios.
Data-Driven Menstrual Analysis of Stress and Mood in Indian Women
Surveyed 219 Indian women to analyze how life stages affect stress and mood across menarche, menstruation, and menopause phases. Found stress significantly influenced by age, weight, and occupation.
Enhancing Security and Privacy Implications in 5G Network Slicing
Examined 5G network slicing security landscape. Proposed advanced resource management and highlighted security/privacy concerns, offering a robust theoretical framework for secure service-specific networks.
Projects
Automated DevOps CI/CD Pipeline for ServiceNow
Architected a robust CI/CD pipeline using GitHub Actions across four ServiceNow instances, replacing bi-weekly manual deployments with an agile, on-demand delivery model. Integrated ServiceNow's Automated Test Framework (ATF) for pre-deployment validation, and built a reporting engine that auto-generates and disseminates release notes to stakeholders — eliminating manual overhead and establishing a transparent, self-service deployment culture.
Multimodal Neurophysiological Framework for Cognitive Behavior Analysis
Built a multimodal neural network for a ₹2 Crore National Government Grant in collaboration with DRDO INMAS & CDAC for National Defense applications. Developed a real-time stress and lie detector fusing EEG, EOG, ECG, Audio, Video, Gaze, and GSR modalities via multi-fusion methods on primary collected data. Achieved 94% accuracy through feature engineering that transforms raw sensor data into meaningful representations.
Agentic AI for Supply Chain Customer Processes
Connecting and automating supply chain processes to improve process handling by 25–40%. Building systems to process data from multiple APIs, from email initiation to case resolution at Maersk.
AIoT Sustainable Health Solution for Wind Turbines
UN SDG funded grant collaboration with UAE University. AIoT-based smart sensor approach for reliable predictive maintenance incorporating multivariate sensor data for wind turbine blades.
Bias Mitigation in Facial Emotion Recognition
Improving robustness of FER models by mitigating 'good-image' bias with synthetic images on the SFEW dataset to enhance security model performance. Published at FG2024, Istanbul.
Deep Learning Medical Image Segmentation with XAI
Collaboration with Aston University, UK. Compared DL-based segmentation algorithms on breast cancer ultrasound images using Grad-CAM for explainability. Presented at GGP-PAAI.
Traffic Light Optimization with Deep Q-Learning
Reinforcement learning to optimize traffic signal control at intersections, reducing congestion and vehicle wait times by adapting to real-time traffic density conditions.
IntelliML — Automated Machine Learning Framework
End-to-end automated ML solution for numerical datasets with auto pre-processing, EDA, model selection, and relevant metrics — reducing time-to-insight dramatically.
SymbiGate — Facial Recognition Attendance System
A smart attendance system built in Java using facial recognition. Captures and matches faces against a database in real time via OpenCV, with a JavaFX GUI and MySQL backend for seamless attendance tracking and management.
Transformer-Based Technical Interview Assistant
NLP-powered tool that assesses technical interview responses, classifies answers, and provides tailored feedback using transformer-based models.
Medium Articles
Videos
Talks, tutorials & appearances across channels.
From System Prompt to Production: Build an AI Code Reviewer | Aadith Sukumar
Aadith SukumarBuild a Gradio chatbot with the Groq API and explore real-world prompt engineering applications across industries.
Watch on YouTube
Adversarial Debiasing in ML Models Using Synthetic Data | Aadith Sukumar | IdentityShield Summit ‘25
miniOrangeLearn how adversarial debiasing and synthetic data improve machine learning fairness and security in cybersecurity applications.
Watch on YouTube
Advanced Prompt Engineering: Chain of Thought & AI Jailbreaks | Aadith Sukumar
Aadith SukumarLearn advanced prompt engineering, AI security, and prompt injection by exploring reasoning techniques and testing LLM vulnerabilities with Lakera's Gandalf challenge.
Watch on YouTube
LLM Fundamentals & Prompt Engineering Explained (By Building a Game) - Part 1 | Aadith Sukumar
Aadith SukumarLearn how LLMs work, prompt engineering fundamentals, and AI reasoning to build smarter applications instead of relying on trial-and-error prompting.
Watch on YouTube
The Only Free AI/ML Roadmap You Need in 2026 (Part 2) | Aadith Sukumar
Aadith SukumarExplore a free AI roadmap covering machine learning, deep learning, generative AI, and agentic workflows with curated learning resources.
Watch on YouTube
The Only Free AI/ML Roadmap You Need in 2026 (Part 1) | Aadith Sukumar
Aadith SukumarLearn AI and machine learning fundamentals through a free, structured roadmap covering math, machine learning, and industry-ready AI skills.
Watch on YouTubeContact Me
Open to collaborations, research discussions, and new opportunities. Feel free to reach out.