Amudhan Manimaran
ML Researcher
About Me
I am Amudhan Manimaran, a third-year B.Tech student specializing in Artificial Intelligence and Machine Learning at SRMIST - Kattankulathur Campus (CGPA: 8.87/10). My research is centered on unsupervised machine learning, computer vision, and principled evaluation methodology — areas where I believe mathematical rigor and real-world applicability must go hand in hand. My research journey began at NIT Tiruchirappalli, where I worked as a Research Intern on affective computing - developing an unsupervised pipeline to model emotional responses in visual art through color science and psychological theory. This work led to the derivation of a novel evaluation metric, Weighted Emotion Accuracy (WEA), achieving 90.3% accuracy in ablation studies. Two first-authored papers from this research are currently under review at Signal, Image and Video Processing, Springer Nature. A third paper on hierarchical plant disease detection using MobileNetV2 has been accepted at ICCET 2026 (Scopus Indexed) and is in press in a Springer Nature edited volume on AI in Agriculture. Beyond research, I ranked in the Top 5% (NPTEL Topper) among 9,715 candidates in Introduction to Machine Learning at IIT Kharagpur, and was selected among the Top 80 candidates at SRMIST KTR for Samsung Innovation Campus - a nationally competitive industry training program under Samsung R&D Bangalore. I build ML systems that span healthcare diagnostics, agricultural intelligence, and renewable energy modeling - always prioritizing reproducibility, mathematical transparency, and research-grade evaluation. I am actively seeking research internship opportunities where I can contribute to high-impact ML research under academic mentorship.
AI-ML STUDENT
"It's not who I am underneath, but what I do that defines me"
- ♠ Birthday: 11 February 2006
- ♠ City: Mayiladuthurai, Tamil Nadu
- ♠ Languages Known: Tamil, English, Telugu, Hindi
- ♠ Research Interests: Unsupervised ML, Computer Vision, Affective Computing
- ♠ Publications: 2 × Springer Nature (Under Review), 1 × Scopus Accepted
- ♠ University: SRM Institute of Science and Technology
- ♠ Degree: Bachelor of Technology
- ♠ Branch: CSE (AI & ML)
- ♠ CGPA: 8.87
- ♠ 12th Board: 85.5%
- ♠ Strengths: Problem Solving, Critical Thinking
My broader project portfolio spans healthcare diagnostics, agricultural intelligence, and renewable energy modeling. Key works include a Tri-Level Quantum-Classical CT Diagnostic Pipeline achieving 95.31% accuracy across 4,613 scans, and an AgroVision plant disease detection system achieving 96.4% accuracy — both built with a focus on deployment-readiness and reproducible evaluation. I am currently a Samsung Innovation Campus Trainee (Top 80 at SRMIST KTR), advancing algorithmic problem-solving skills under Samsung R&D Bangalore's engineering curriculum. I approach every research problem with mathematical rigor, a commitment to reproducibility, and a drive to build systems that work beyond controlled benchmarks.
Skills
Programming Languages
AI & ML Frameworks
Data & Research Tools
Computer Science Fundamentals
Web Development & Deployment
My Experiences
"Highlighting a journey through specialized industry training, research experiences, and cloud-native data engineering, with a core focus on Artificial Intelligence and Machine Learning."
Research Intern – Affective Computing
NIT Tiruchirappalli (NITT)
- Engineered an unsupervised computational pipeline using Gaussian Mixture Models (K=25) to extract probabilistic chromatic fingerprints from a curated corpus of digital art.
- Formulated a deterministic 48-class affective feature space mapping Plutchik's taxonomy to RGB coordinates, ensuring geometric coherence across the emotion manifold.
- Derived a novel Weighted Emotion Accuracy (WEA) metric, achieving 90.3% accuracy in ablation studies and a +0.161 absolute gain over traditional unsupervised baselines.
Samsung Innovation Campus Trainee
Samsung R&D Bangalore (SRIB)
- Selected as one of the top 80 candidates from a nationally competitive cohort at SRMIST KTR for Samsung’s flagship Coding & Programming program.
- Trained under SRIB’s corporate curriculum on industry-grade infrastructure, gaining exposure to rigorous software engineering standards and research workflows.
- Optimizing complex algorithms across Graphs, Dynamic Programming, and Greedy heuristics under strict technical constraints benchmarked by Samsung R&D.
Data Engineering Intern
AWS (via Eduskills Foundation)
- Architected cloud-native ETL processes and streamlined data pipelines to handle high-velocity datasets using AWS Glue and Lambda.
- Implemented structured storage solutions in S3 and Redshift, focusing on data normalization and query performance optimization.
- Successfully achieved AWS-specific certifications while applying theoretical knowledge to simulated enterprise data scenarios.
AI-ML Intern
Google (via Eduskills Foundation)
- Executed extensive exploratory data analysis (EDA) and preprocessing for diverse datasets to ensure high-fidelity model training.
- Applied and tuned ML algorithms including Random Forests, K-Means, and Ridge Regression to solve predictive modeling challenges.
- Focused on rigorous model evaluation metrics to bridge the gap between sandbox training and real-world inference accuracy.
Analytics Process Automation Intern
Alteryx (via Eduskills Foundation)
- Designed and deployed automated data workflows for complex analytical tasks, significantly reducing manual processing latency.
- Leveraged the Alteryx Sparked curriculum to perform advanced data blending and spatial analytics for location-based intelligence.
- Integrated end-to-end pipelines that serve as the foundational layer for scalable predictive modeling and business intelligence.
My Projects
"Explore my collection of projects that demonstrate my proficiency in AI, ML, computer vision, full-stack development, and data science, tackling real-world problems with robust and innovative tech solutions."
ChromaSense – Unsupervised Affective Art Analysis Pipeline
- Formulated a computational affect modeling framework during a research internship at NIT Trichy, bridging color science and Plutchik's Wheel of Emotions to formalize subjective visual data into 48 quantifiable emotional states.
- Architected a dual-clustering pipeline utilizing K-Means and Gaussian Mixture Models (GMM) to extract dominant perceptual color centroids, mapping high-dimensional RGB features to discrete emotional categories via Euclidean distance modeling.
- Curated a structured 48-class psychophysical RGB dataset to serve as a synthetic ground truth, and built an interactive Flask dashboard to visualize HSI distributions and validate inferences using a novel Weighted Emotion Accuracy (WEA) metric exceeding 90%.
AgroVision - Hierarchical Plant Disease Detection System
- Designed a two-stage cascaded MobileNetV2 pipeline with conditional logic routing for species-specific disease classification, achieving 96.4% accuracy with a lightweight, deployment-ready footprint.
- Leveraged inverted residual blocks for low-latency inference, making the model suitable for resource-constrained field environments.
- Built a real-time Flask dashboard for instantaneous diagnostic visualization and decision support.
Multi-Organ CT Diagnostic Pipeline - Tri-Level Quantum-Classical Architecture
- Architected an automated three-tier hierarchical pipeline for organ routing, anomaly pre-screening, and pathology classification, achieving 95.31% overall accuracy across 4,613 hold-out scans.
- Implemented a Hybrid Quantum-Classical CNN integrating a fine-tuned ResNet-18 backbone with an 8-qubit Variational Quantum Circuit, achieving 97.44% accuracy in multi-class kidney pathology detection.
- Prototyped the complete seven-model framework as a locally deployed clinical decision-support system featuring a REST API, structured diagnostic dashboard, and automated PDF reporting.
AeroHeal - Autonomous Server Remediation via Reinforcement Learning
- Architected an end-to-end self-healing infrastructure pipeline using PPO to autonomously monitor 4-dimensional server telemetry (CPU, Memory, Latency) and execute proactive remediation before system failure.
- Devised a custom simulation environment to process continuous streams of stochastic system data, optimizing the system to prioritize proactive intervention over reactive crash recovery.
- Optimized a high-throughput processing architecture to evaluate 300,000 steps of system telemetry, maintaining strict SLA constraints without manual threshold engineering.
Healthcare Anomaly Detector - Multivariate Safety Monitoring
- Engineered an XGBoost anomaly detection model on high-dimensional healthcare data to flag phantom billing and diagnostic errors, supporting health system planning.
- Handled severe class imbalance using SMOTE, achieving a 99% F1-score in distinguishing critical anomalies from benign outliers.
- Built a reproducible real-time alerting dashboard (Flask) to surface deviation metrics for safety-critical monitoring.
HelioCast - Solar Cell Efficiency Prediction via Polynomial Ridge Regression
- Developed a Polynomial Ridge Regression model from scratch (without high-level ML libraries) to predict solar cell efficiency, ensuring full mathematical transparency.
- Modeled non-linear energy generation curves to identify efficiency bottlenecks and optimize output prediction for renewable energy systems.
- Demonstrated white-box interpretability, making the model suitable for techno-economic analysis in industrial R&D settings.
GlucoGuard – Intelligent Diabetes Prediction Platform
- Engineered a predictive machine learning model using Scikit-learn to assess diabetes risk from patient health records, achieving an 85% accuracy and a 0.86 macro F1-score.
- Executed robust data preprocessing pipelines, including feature normalization and class balancing, to significantly optimize precision and recall rates for positive clinical cases.
- Prototyped a real-time Flask web application to surface actionable diagnostic insights, categorizing patient profiles into Low, Moderate, and High-risk tiers.
GesturePlay – Real-Time Visual Processing & Logic Engine
- Designed an interactive framework that processes live visual inputs and explicitly enforces strict, predefined rules, demonstrating the ability to manage complex, constraint-driven workflows.
- Implemented a state-aware decision engine to evaluate dynamic inputs and execute valid actions under strict parameters, showcasing strong analytical and rule-satisfaction capabilities.
- Optimized a real-time visual processing loop using OpenCV to ensure seamless, low-latency performance while maintaining rigorous adherence to systemic logic during execution.
ScriptScout- LLM Powered Manuscript Compliance Checker
- Integrated Gemini API with structured prompt engineering to automate manuscript auditing against IRJMS compliance rules, replacing manual editorial screening.
- Crafted constraint-based prompts to decompose policy documents into checkable validation criteria, enabling Gemini to flag violations and generate improvement recommendations.
- Constructed a PDF/DOCX text extraction and normalization layer to feed clean, structured manuscript content into the Gemini pipeline for reliable compliance checks.
CareerBridge - Agentic Guidance & Skill Pathway Ecosystem
- Architected an end-to-end agentic guidance framework integrating a high-precision diagnostic scanner with a context-aware conversational agent to deliver personalized career trajectory consulting and real-time feedback.
- Engineered a persistent state management layer utilizing a Seq2Seq Transformer architecture, enabling cross-turn context retention and dynamic skill-gap remediation over static analysis.
- Designed a hybrid recommendation engine that maps identified competency gaps to targeted educational resources and interview strategies, replacing manual screening with an outcome-oriented guidance model.
University Management System – Full-Stack Administration Portal
- Developed a Java-based full-stack web application with robust CRUD operations to streamline administrative workflows for students, faculty, and course management.
- Integrated a normalized relational SQL database with optimized queries to ensure scalable, efficient, and reliable enterprise data handling.
- Implemented secure authentication and role-based access control (RBAC) protocols to enforce stringent data governance.
Cricketers Stats Management System
- Built a full-stack web application using Flask and MySQL to efficiently manage, update, and explore complex relational statistics.
- Constructed interactive frontend templates featuring real-time search functionality and automated data computation for streamlined user insights.
- Devised secure server-side validation protocols and a tabbed, user-friendly frontend interface to ensure safe data entry and management.
Certifications
"A testament to my commitment to continuous learning and professional growth, these certifications showcase my expertise and dedication to excelling in the fields of AI, machine learning, and software development."
