Senior Machine Learning Engineer

• Mar 2022 - Present

Machine Learning Operations Engineer

• Mar 2022 - Dec 2022

One of the biggest successes I achieved on this job was to reduce the Model Training time from 108 hours to 7 seconds. Building machine learning models at scale and deploying them (following Docker containerization, creating microservices architecture and Kubernetes as a managed service). The unique_name in the dataset to identify the sales of a product would change every year so there was no way to figure out how many sales one specific product has had over the last couple of years and predict the sales for them. As a solution, I proposed and built a BERTForSequenceClassification model that would find the products similar as per sales with 95% accuracy, reducing the time to do data annotation for 200 million records and hence saving the company $50,000 per annum. To predict the sales through text-based data, built a regression model through the former transformer-based model where in a unique way, all the features (text+numerical) were concatenated, tokenized, and then passed to build the model and got extremely interesting results. (This approach exponentially reduced the model training time and increased the model accuracy from 78% to 93%. Built similarity-based models using transformers and trained on multiple GPUS using distributed (Techilla instances).Built a model around the Statistics Canada dataset i.e. targetting specific ethnicities in specific cities of Canada while doing sales and helped the model predict the sales per door code in each city through the presence of each ethnicity at each zip code. Show less

Teaching at Data Analytics and Visualization Boot Camp (Virtual)

• Jun 2021 - Present

Teaching an intensive 24-week Data Analytics and Visualization Bootcamp. • Provided insight, support, creative solutions, and project management aid to individuals and student teams to accomplish coding tasks and projects using a multitude of technologies including Python, Pandas, Pipenv, Jupyter, Git and Github, CLIs, R, sci-kit-learn, Flask, Plotly, Tableau, HTML, CSS, Bootstrap, NoSQL and SQL databases, JavaScript, D3, API’s and CSV’s. • Resulted in longstanding collaborative relationships with graduates, including increased enrollment via student and personal referrals, and the highest NPS across coding boot camps at McCombs. • Lecturered for 3+ Classes as a substitute instructor More Details about the use of technology: • Gathered data from relational databases, NoSQL databases, APIs and flat files using MySQL, MongoDB, Python and Pandas • Performed exploratory data analysis and modeling using MS Excel, R, Python, Pandas, Matplotlib and scikit-learn • Built APIs and web back-ends using Python and Flask • Used JavaScript, D3.js, Plotly.js, Leaflet.js, HTML, and CSS to make web front-ends, visualizations and dashboards • Utilized Git and GitHub for collaboration on coding projects • Managed dependencies and virtual environments with pip, pipenv, and poetry Show less

Instructor

• Dec 2021 - Present

Teaching a complete End to End MIT Based Data Science Bootcamp to students on Python and Statistics for building data science applications. Performing Exploratory Data Analysis for extracting insights and then making business decisions based on the key features. Analyzing Graphs and Network-based analysis in the datasets, Building Classification and Regression-based models including (clustering, Ensemble learning, and Bootsrap-based models), Time Series Analysis, and Model Evaluation for different ML use cases. Building Deep Learning models on audio and text analysis, recommendation systems, and Transfer Learning. Finally, teaching the students about Model Deployment and building machine learning models at scale. Show less

Teaching Assistant

• Jun 2022 - Dec 2022

Machine Learning Developer

• Apr 2021 - Mar 2022

Instrumental in researching, prototyping, designing, implementing and evaluating machine teaming models • Digging deep into millions of data records and performing data exploratory analysis using interactive Jupyter notebooks using pandas/numpy/matplotlib • Researched, prototyped (from research papers), built features and optimized(hyper-parameter tuning) the state-of-the-art machine learning and deep learning techniques like SVM, Logistic Regression, Random Forest regression, LSTM, CNN etc.. using scikit-Learn, keras, tensorflow • Containerized (Docker) and Deployed the re-producable machine learning models into production through Jenkins Cl/CD orchestration onto AWS ECR Repositories/S3-artifacts/k8s after code reviews in an agile process. • Automated the data science platform by building and managing data pipelines and extensions that bridges NoSQL-databases, APIs, Machine Learning Engine and tits. • Applied various transfer-learning techniques using pre-trained word-embeddings like Glove. fastText, Glove. BERT, ELmo, • Post Training Quantization using Tensorflow-lite: Deep learning models that get executed on GPU only were quantized to make them able to run on CPU by reducing the size of the model and increasing their inference speed. • WhyNot Model: Finding the features that led to the prediction of a machine learning model to be a NO and why would the model say NO by finding the respective weights for each feature. • Documents Clustering: Millions of documents were passed through an unsupervised learning model using fastText to get the documents that are similar to each other and finding the relevancy between those clusters. • Transformers model: Implementation of a Transformers model using hugging-face and spacy to get the correlated similar sentences and finding the sentences from a corpus based on the given context. Show less

Data Scientist

• Jan 2021 - Apr 2021

Building End to End Machine Learning projects with deployment using 𝗗𝗼𝗰𝗸𝗲𝗿, 𝗞𝘂𝗯𝗲𝗿𝗻𝗲𝘁𝗲𝘀 and 𝗛𝗲𝗿𝗼𝗸𝘂. Few of the projects currently in progress are below: 𝟭) 𝗨𝗻𝗶𝘁𝗲𝗱 𝗡𝗮𝘁𝗶𝗼𝗻𝘀 𝗵𝗮𝗽𝗽𝗶𝗻𝗲𝘀𝘀 𝗶𝗻𝗱𝗲𝘅 (𝗣𝗿𝗲𝗱𝗶𝗰𝘁𝗶𝗼𝗻) According to UN report, 6 socio-economic feature i.e. health, family support, economic strength, corruption rate determine the happiness of a state. The data is important for the policy development institutes so a notebook is deployed to see the essence of those factors. 𝗚𝗶𝘁𝗵𝘂𝗯 𝗖𝗼𝗱𝗲 𝗨𝗥𝗟: 𝗵𝘁𝘁𝗽𝘀://𝗴𝗶𝘁𝗵𝘂𝗯.𝗰𝗼𝗺/𝗮𝘀𝗶𝗺𝘀𝘂𝗹𝘁𝗮𝗻/𝗨𝗡-𝗛𝗮𝗽𝗽𝗶𝗻𝗲𝘀𝘀-𝗥𝗲𝗽𝗼𝗿𝘁-𝗔𝗻𝗮𝗹𝘆𝘀𝗶𝘀 𝟮) 𝗣𝗮𝘁𝘁𝗲𝗿𝗻 𝗥𝗲𝗰𝗼𝗴𝗻𝗶𝘁𝗶𝗼𝗻 𝗶𝗻 𝗛𝘂𝗺𝗮𝗻 𝗗𝗡𝗔: In order to find specific set of human classes on the basis of their DNAs (A long sequence of characters that has no explicit meaningful information), 𝗞-𝗺𝗲𝗿𝘀 𝗰𝗼𝘂𝗻𝘁 𝗮𝗻𝗮𝗹𝘆𝘀𝗶𝘀 algorithm is being used along with 𝗠𝘂𝗹𝘁𝗶𝗻𝗼𝗺𝗶𝗮𝗹 𝗡𝗮𝗶𝘃𝗲 𝗕𝗮𝘆𝗲𝘀 Classifier to predict the hidden patterns in the sequences. 𝗚𝗶𝘁𝗵𝘂𝗯 𝗖𝗼𝗱𝗲 𝗨𝗥𝗟: 𝗵𝘁𝘁𝗽𝘀://𝗴𝗶𝘁𝗵𝘂𝗯.𝗰𝗼𝗺/𝗮𝘀𝗶𝗺𝘀𝘂𝗹𝘁𝗮𝗻/𝗗𝗡𝗔-𝗣𝗮𝘁𝘁𝗲𝗿𝗻-𝗥𝗲𝗰𝗼𝗴𝗻𝗶𝘁𝗶𝗼𝗻 𝟯) 𝗣𝗮𝘁𝗶𝗲𝗻𝘁𝘀 𝗦𝘁𝗿𝗼𝗸𝗲 𝗣𝗿𝗲𝗱𝗶𝗰𝘁𝗶𝗼𝗻 Stroke is the 2nd leading cause of death globally, responsible for approximately 11% of total deaths, A project is deployed to predict this disease for every human based on the real dataset. 𝗗𝗲𝗺𝗼: https://strokepredictioninpatients.herokuapp.com/ 4) 𝗨𝗻𝗶𝘃𝗲𝗿𝘀𝗶𝘁𝘆 𝗔𝗱𝗺𝗶𝘀𝘀𝗶𝗼𝗻 𝗣𝗿𝗼𝘁𝗼𝗰𝗼𝗹 𝗣𝗿𝗲𝗱𝗶𝗰𝘁𝗶𝗼𝗻: The given project is to predict the probability of a student getting admission in a university. 𝗗𝗲𝗺𝗼 : 𝗵𝘁𝘁𝗽𝘀://𝘀𝘁𝘂𝗱𝗲𝗻𝘁𝗮𝗱𝗺𝗶𝘀𝘀𝗶𝗼𝗻𝗽𝗿𝗲𝗱𝗶𝗰𝘁𝗶𝗼𝗻-𝗮𝗽𝗶.𝗵𝗲𝗿𝗼𝗸𝘂𝗮𝗽𝗽.𝗰𝗼𝗺/ Show less

Data Scientist

• Jun 2020 - Mar 2021

Machine Learning Engineer (Python Developer) • Designed a complete system for Data Engineering team for 1 million customers, writing efficient data pipelines to enrich the utilization of programs into memory. • Predicted the revenue of the company with 91% accuracy to enable them to dispatch calls with potential Customers. • Developed an algorithm to understand effect of weather on the number of calls received, leading to 35% Fuel savings and calling only a specific number of drivers to work • Boosted the company`s revenue by 13.7% annually, developing an Autoregressive Integrated Moving Average (ARIMA) time series algorithm to analyze the dispatched calls movement with respect to given transactions • Devised high-performance ML systems to detect abnormality, Outliers and any irregular transaction using PCA and Isolation Forest • Assisted in model development, testing, and validation of Automotive products and services • Created charts in Jupyter Notebook to perform analysis & visualize data using Matplotlib • Developed Restful APIs using Django Rest Framework Show less

Machine Learning Specialist

• Jun 2017 - Sep 2019

Machine Learning Developer • Developed customer segmentation algorithm using PCA leading to 22% increase in the Classification task • Optimized personalization algorithms for applications with 2M+ users • Applied data mining to shipping consolidation problem, saving $1.2M • Predicted product sales to within 2% by applying logistic regression model Data Analyst/NLP Engineer • Designed and developed analysis systems to extract information from large scale data created social media sentiment analyzer that tracks 150 million posts per day • Implemented an NLP research paper Double Propagation Algorithm to detect the opinions with in the reviews and target the specific entities within the text using Dependency Parsing in Python Show less

Machine Learning Intern

• Jun 2016 - Aug 2017

Data analytics such as building classifiers for Machine learning algorithms and text processing systems using Natural Language Processing. Naive Bayes Classifier, SVM, KNN, Perceptron etc are the classifiers that I have worked on. Classification, Clustering, Regression using Hadoop and HDFS systems. Spark environment in python (pyspark) for extracting live streaming data and then performing sentiment analysis. Reviews of the amazon site has been extracted to make a recommendation system for a particular product. Text Analytics, Document summarizing, Polarity of the extracted features from the data. Extracting the Patterns from the Big data system, in particular for the medical systems. Show less

Mobile Application Developer

• Jan 2016 - Sep 2016

An advanced replica of Whatsapp application was developed using state of the art Real time Database (Google Firebase) with an advancement of real time messaging. The application performs all the functions as Whatsapp but has two more advanced features. 1: The message sending speed is realtime means it takes less time to send message to a user through this app than on whatsapp. 2: Whatsapp would only show the users in its contact directory that are saved on the respective device whereas our application would also show those contacts who have saved this user`s contact but this user doesn`t know have him saved on this device. Show less