Data Scientist - ML Engineer - AI Researcher

HEMAND ADISH RAMESHKUMAR

Myportfolio # ********@********.*** ï hemand-adish § GitHub

EDUCATION

University of Colorado Boulder Aug 2024 – May 2026 Master of Science in Data Science Boulder, CO

Sri Krishna College of Engineering and Technology Aug 2019 – Apr 2023 Bachelor of Engineering, Mechatronics Engineering Tamilnadu, India SKILLS AND INTERESTS

Languages: Python, Hadoop, R&, SQL, Data warehouse, Data visualization, PowerShell, Qlik, Power BI, MongoDB, Databricks, NoSQL, Scala, Cloud infrastructure (Azure), Snowflake, Spark, Product ecosystem, Agile Tools and frameworks: Tableau, Tensorflow, Pytorch, Kafka, Apache, Hive, SAP, WebKit, data architecture Interests and skills: Data Scientist, Data analytics, Business Analytics, Database, Business Intelligence, ETL, ML

(predictive models), AI, Data structures, Generative AI, Data Mining WORK EXPERIENCE

Research Intern(Patent) – AI & Machine Learning May 2022 – Jun 2023 Sri Krishna College of Engineering and Technology Tamilnadu, India

• Built ML models (XGBoost, RF) for predictive maintenance using real-time CAN bus data; achieved 87% accuracy.

• Designed Kafka + SQL pipeline for V2V anomaly detection; reduced false positives by 40%.

• Deployed Dash + Flask dashboard for real-time diagnostics; cut response time by 30% and contributed to a patent on edge AI fault detection using embedded ML and vehicle telemetry. Data Analyst and Deep Learning, Internship Oct 2021 – Mar 2022 DeepVisionTech.AI Bengaluru, India

• Established web scraping pipelines, achieving 95-99% by extracting features and maintaining an error rate below 5%, ensuring high-quality data inputs for AI models.

• Enhanced NLP preprocessing to over 95%, boosting model F1 scores to 0.8, optimizing input handling for better machine learning performance.

• Created a machine learning pipeline with a custom ANN, achieving 98.79% precision and reducing computation time by 20%. Applied TensorFlow and dimensionality reduction methods. PATENTS & PROJECTS

PairWiseRL: RL-Based Two-Item Product Bundling (under review for Journal) Apr 2025 – Jun 2025

• Designed and implemented a custom OpenAI Gym environment and Deep Q-Network agent to learn optimal two-item bundles from the RetailRocket e-commerce dataset using reward shaping (in-basket + future-purchase signals) and seeded exploration.

• Achieved a >20 percent improvement in single-item hit rate over a random baseline within 50 training episodes, demonstrating effective, self-tuning bundle recommendations. Self-Supervised Learning for Tabular Data Inspired by NLP and Vision Nov 2024 – Jan 2025

• Built a hybrid SSL framework using Masked Feature Modeling and Contrastive Learning, adapting techniques from masked language modeling and instance discrimination for structured tabular data.

• Outperformed supervised baselines on MIMIC-III, UCI, and IEEE-CIS datasets using PyTorch, SHAP, and UMAP; enabled applications in fraud detection, risk prediction, and churn modeling. ChatBot by using RASA open-Source and RASA-X Jun 2021 – Aug 2021

• Constructed an AI chatbot employing Rasa (NLU, Core, Rasa-X) on a VM instance, achieving 90%+ intent classification accuracy for real-time COVID-19 information. Link

• Optimized NLP pipelines, reducing response time by 40% and improving user query resolution efficiency. PATENTS AND PAPER PUBLICATIONS (chronological)

Patent- AI/ML SystemforReal-TimeFaultDetectioninHybridVehicles Link Published a patent in Germany Jun 2023, App.no: (DE202023101526) Paper- Robotics Innovation withReal-TimeMLControl Link International Conference Innovations in Robotics Oct 2022, Intelligent Automation and Control

Contact this candidate