Aniket Phutane Email & Phone Number

Berlin, Germany

Aachen, North Rhine-Westphalia, Germany

• Developed a Multi-agent Reinforcement Learning solution for topology optimization of power grid networks, resulting in more efficient, and resilient power grid networks.• Implemented actor and critic networks utilizing Transformer architecture - Multi-headed attention networks with Positional Embedding to enhance learning capabilities.• Utilized Python, with libraries such as PyTorch, Pandas, and NumPy, for model development and analysis.• Conducted A/B testing on variations of the transformer architecture and different reward functions to identify the optimal hyper-parameters.• Managed version control and collaborative development with GitLab, ensuring efficient teamwork.• Employed High-Performance Computing (HPC) clusters for model training, enhancing computational efficiency.

Ludwigshafen, Rhineland-Palatinate, Germany

Biodegradability Prediction of Small Molecules• Applied Graph Convolutional Networks (GCN) and contrastive learning to learn a representation space for a large unlabeled dataset.• Utilized features derived from representation learning in a Random Forest model for the prediction task due to limited labelled data.• Conducted A/B testing to compare the performance of deployed models and fine-tuned the algorithms for optimal results.• Collaborated with Chemists and Biostatisticians to verify and test experiment results, ensuring the accuracy and reliability of findings throughout the development phase.Explainability of Model Predictions• Ensured explainability of model predictions by using SHAP (SHapley Additive exPlanations) to understand the contribution of features to each predicted score. • Used the Random Forest algorithm to generate feature importance scores, created Partial Dependence Plots to reveal the relationship between features and predicted outcomes.• Conducted consistency checks to ensure that model predictions align with domain knowledge.Tech Stack: Python (Pytorch, Pandas, Numpy, SKLearn, Statsmodels), SHAP, Docker, MLflow, DVC, Gitlab, Flask.

Aachen, North Rhine-Westphalia, Germany

Predictive Maintenance of Aircraft Turbo Engines• Managed a PostgreSQL database to efficiently store and retrieve time-series sensor data in Python using Psycopg2.• Developed a CNN-1D forecasting model with a 7% error rate in validation against ground truth data. • Implemented Variational Autoencoders (VAE) for dimensionality reduction and deployed the model seamlessly on AWS, ensuring streamlined versioning and tracking.Anomaly Detection in Transformers• Implemented an Isolation Forest model to detect anomalies in time-series Transformers data.• Conducted feature selection using p-value, forward and backward selection techniques.• Developed a real-time Alert notification system with Flask, triggering automated email alerts when values surpassed predefined thresholds. Tech Stack: PostgreSQL, AWS (Sagemaker, Lambda, S3, API Gateway), Python (Flask, Pytorch, Keras, Pandas, Numpy, SKLearn, Psycopg2. Statsmodels), Gitlab, Docker, Kubernetes, Grafana.

• Performed real-time analysis on datasets using Celonis Process Explorer and recognized the optimal process flow to reduce cost by 4%.• Identified the bottom 20 percentile cumulative spend of 45 entities using Structured Query Language on SAP-HANA.• Implemented a Python pipeline and automated scheduled triggering of daily email narratives consisting of the Key Performance Indicators (KPIs).

Mumbai, India

• Examined and implemented - Random Forest for feature selection and time series forecasting method –Seasonal ARIMA for predictive analytics.• Reduced the Mean Absolute Percentage Error to 12% from the initial 25%.• Translated customer data insights into business recommendations by creating 20+ Tableau dashboards with relevant stakeholders.

Mumbai Area, India

• Primarily performed Data Analysis and Anomaly Detection for Energy-based Datasets.• Implemented K-means algorithm for data clustering followed by Isolation Forest for outlier detection to identify sensor based anomalies.

Master Of Science - Ms, Data Science, 2.3

Relevant Coursework: Automatic Speech Recognition, Advanced Machine Learning, Mathematics of Data Science, Computer Vision and Applied.

Bachelor Of Engineering, Computer Engineering, 8.26

Relevant Coursework: Machine Learning, Soft Computing, Algorithms and Data Structures, Applied Mathematics

Higher Secondary Education, Science

Gold Medalist at the International Olympiad of Mathematics, 2011.