Avishek Paul Email & Phone Number

Montreal, Qc, Ca

Montreal, Quebec, Canada

• Received CEO Award for development of an internal tool that generates a Requirement Tracking Matrix that maps each customer requirement to OSI product test procedure. Mapping is done through semantic similarity, yields high accuracy and reduce manual efforts of engineer by 60%• Received People's Choice Award at AspenTech Tech Summit 2023 for developing an AI powered Visualization tool for WebStudio that automatically determines the best KPI metrics, plots, analysis summary and… Show more • Received CEO Award for development of an internal tool that generates a Requirement Tracking Matrix that maps each customer requirement to OSI product test procedure. Mapping is done through semantic similarity, yields high accuracy and reduce manual efforts of engineer by 60%• Received People's Choice Award at AspenTech Tech Summit 2023 for developing an AI powered Visualization tool for WebStudio that automatically determines the best KPI metrics, plots, analysis summary and a chat mechanism from a given datset. It provides an improved analytical tool for user• Developed a Graphical User Interface to query and generate sections of Functional Design Specification (FDS) document by performing Retrival Augmentation Generation on previous sample FDS documents. Adoption of this technique can drastically reduce time needed for document preparation by 70%• Training agents with reinforcement learning to alleviate congestion on power systems network using topological actions (line and bus switching) and suggest possible set of sequential actions to operators• Contributing to improvement of internal Time Series Forecast library that compares various statistical/deep-learning models and presents best one after hyperparameter search. Incorporating SOTA models (eg. Chronos-TS) into library.• Leveraging state of the art machine learning techniques to develop simplified applications for users of existing software Show less

Montreal, Quebec, Canada

• Resolving software issues, bugs and feature enhancement as requested by customers of Energy Management System softwares ( OpenNet, OpenVSA etc.)• Triaging and debugging issues in C language and working in an agile environment with 2 week sprint planning• Communicating with Project Delivery and other development teams and writing development documentation and implementing new features.• Testing new features on test system before releaseSoftware used : Visual Studio (C),… Show more • Resolving software issues, bugs and feature enhancement as requested by customers of Energy Management System softwares ( OpenNet, OpenVSA etc.)• Triaging and debugging issues in C language and working in an agile environment with 2 week sprint planning• Communicating with Project Delivery and other development teams and writing development documentation and implementing new features.• Testing new features on test system before releaseSoftware used : Visual Studio (C), Version Control (github, SVN) Work management tools (JIRA) Show less

Montreal, Quebec, Canada

• Quantizing pretrained models (MobileNet/ InceptionNet/ FaceNet) for face detection through (a) Post Training Static Quantization (b) Quantization Aware Training• Comparing and evaluating performance of actual and quantized model using popularly available datasetsSoftware used: pytorch (quantization), Visual Studio Code

Montreal, Quebec, Canada

• Successfully established a pipeline to extract features from a massive dataset of pre-segmented audio segments (called cry units) and save them.• Plotting and segmenting audio signals and spectrograms with provision of saving• Developed process that facilitates ease of performing exploratory data analysis to identify features that readily demarcates the cohorts• • Established a process for development on local machine through jupyter-notebook that access VM instance on Google… Show more • Successfully established a pipeline to extract features from a massive dataset of pre-segmented audio segments (called cry units) and save them.• Plotting and segmenting audio signals and spectrograms with provision of saving• Developed process that facilitates ease of performing exploratory data analysis to identify features that readily demarcates the cohorts• • Established a process for development on local machine through jupyter-notebook that access VM instance on Google Cloud Platform and data on Google StorageSoftwares used: Google Cloud Platform, Colab notebooks, matplotlib, pandas. scikit-learn, crepe Show less

• Understanding the problem objective, modelling and executing Graph Neural networks ( Graph Convolutional Networks, Graph Attention Networks, Graph Classification) for link prediction, text/sentiment classification• Data cleaning and formatting and running pre-established established models on customer datasets Software used : Spyder, jupyter-notebook, google-collab, dgl, torch_geometric, pandas

Montreal, Canada Area

• Compared three different types of classifiers for differentiating zebra finch bird song based on communicative behavior : (a) Decision tree classifier trained on a massive set of extracted features from the songs (b) Long Short Term Memory network classifier with attention mechanism on overlapping segmented sections of songs (c) Convolutional Neural Network on spectrogram images of the songs (https://github.com/AvisP/ZebraFinchClassification)… Show more • Compared three different types of classifiers for differentiating zebra finch bird song based on communicative behavior : (a) Decision tree classifier trained on a massive set of extracted features from the songs (b) Long Short Term Memory network classifier with attention mechanism on overlapping segmented sections of songs (c) Convolutional Neural Network on spectrogram images of the songs (https://github.com/AvisP/ZebraFinchClassification) • Used matplotlib, seaborn networkx to visualize results (a) Low dimension projection of feature set using TSNE (b) UMAP projection of spectrogram of distinct syllables (c) Color coding of individual sequence segment proportional to LSTM prediction (d) Graphs after fitting a Hidden markov model• Leveraged open source algorithms for complex modelling (https://github.com/AvisP/AVGN_Avishek) (a) Dynamic song segmentation (b) Unique syllable identification (c) Transition diagram and Markov chain modelling• Compute graph similarity from first order HMM through Graph Edit Distance to understand degree of variation of song structures among different birds• Study the importance of sequential information ad characterizing vocal learning by Pupils from Tutors in individual nests (a) Feature extraction from spectrograms using states of art CNN networks (DenseNets) (b) Generate sequences using features of individual segments (c) Perform classification using Uni and Bi-directional LSTMs• Interactive fast labelling of syllables in dataset through interactive visuals utilizing bokehSOFTWAREPython (keras, tensorflow, pytorch scikitlearn, matplotlib, seaborn, pandas, pytorch, dgl, torch_geometric ), Matlab Classification Toolbox Show less

Montreal, Canada Area

1) Established a Centralized Dynamic State Estimator (DSE) that deals with communication interruption and delays2) Demonstrated applicability of a multivariate LSTM network for instability prediction using instability indices and contrasted with featured based random forest classifier3) Developed a new Response-based RAS for Instability Mitigation through online identification of runaway generators4) Confirmed code generation of proposed scheme for standalone execution in real time… Show more 1) Established a Centralized Dynamic State Estimator (DSE) that deals with communication interruption and delays2) Demonstrated applicability of a multivariate LSTM network for instability prediction using instability indices and contrasted with featured based random forest classifier3) Developed a new Response-based RAS for Instability Mitigation through online identification of runaway generators4) Confirmed code generation of proposed scheme for standalone execution in real time simulator platforms Softwares Used : Matlab/Simulink, python, HyperSim Show less

Montreal, Canada Area

• Successfully developed a Graphical User Interface (GUI) using Tkinter for easier creation of Wind Farm Layouts in PowerWorld simulator for user• Seamless communication with simulator for automatic data retrieval and analysis using python• GUI generates reports on Equivalent Model, Reactive Power Plots and Loss Optimization and maintains an easily modifiable database of Cables and Wind Turbines for new models

Gurgaon, India

Technology Development( On Study leave)1) Prepared Pollution Map of Northern Region and estimated Equivalent Salt Deposit Density to understand impact on line outages.2) Developed Lightning Map of North-Eastern Region to identify most impacted lines3) Technical Specification of instruments for R&D laboratory prepared for tendering.

Indian Institute Of Technology Delhi

Teaching Assistantship

Doctor Of Philosophy - Phd, Electrical And Computer Science Engineering (Power Systems), 3.9/4.0

Courses taken : Linear Systems, Optimization, Detection and Estimation, Machine Learning Research : 1) Established a Centralized Dynamic.

Master'S Degree, Electrical Engineering Power Systems, A

Activities and Societies: Participated in tech fest TRYST ( cloud computing and ethical hacking workshop) Thesis topic : Wide Area.

Bee, Electrical Enginnering, First Class With Honours

Activities and Societies: Participated in J.U. tech-fest SRIJAN in robotics and FPS gaming events. Courses Taken : Electric Machines.