Ritesh Mishra

I'm a pre-final year student at SRMIST, Kattankulathur, where I have worked and researched mostly in machine learning field along with some experience in web development too.

Email  /  Resume  /  Github

Research

I aspire to become a data scientist at one of the MAANG companies. I am working on that by learning and making unique projects based on Machine Learning and Data Analytics which can be found next.

Engineered a Python script with OAuth for efficient retrieval and organization of code from GitHub repositories. Developed versatile functions to extract code content from diverse file types, including Jupyter Notebooks by using nbformat. Integrated OpenAI's GPT-3.5 Turbo for dynamic, user-friendly analysis and insightful suggestions on provided code snippets.

Developed a paraphrasing model using Word2Vec's n-gram language model trained on 7.8 million sentences (dataset can be found here). The model made was evaluated using a standard word pair similarity dataset called wordsim353, achieving a correlation accuracy of over 65%. It makes use of Python modules such as "re" to preprocess data, NLTK to tokenize data, and Gensim to utilize Word2Vec.

Trained a model on the ISL dataset using LSTM neural networks to detect Indian Sign Language gestures, emphasizing temporal dependencies. Integrated MediaPipe and OpenCV for precise hand landmark detection and real-time visualization during the modeling process. Developed an efficient application for real-time gesture recognition, enhancing accessibility with dynamic visualizations based on the trained model.

Developed a web application for multi-class classification, specifically predicting wine variety using Scikit-Learn algorithms and TensorFlow, achieving more than 99% accuracy with a deep learning model using Word Embedding and LSTM. Implemented the model in a website using Flask and HTML-CSS, and deployed it on Microsoft Azure. Also built an API using Flask for predictions.

Developed a model using TensorFlow to predict the death age of individuals, intended for use in hospitals to generate reports. Trained the model by creating a sequential model, identifying the problem as a regression task, and selecting appropriate loss functions and activation functions based on the dataset. Achieved low Mean Squared Error and loss function values of 14 and 12, respectively, for accurate predictions based on the given dataset.

Developed a model using Python libraries like Pandas, Numpy, Sci-Kit Learn, and Seaborn to predict the least crowded time at the gym, aiding people in scheduling efficiently. Preprocessed the dataset and performed exploratory data analysis to determine suitable algorithms for the dataset. Tested multiple algorithms and employed methods like feature selection to enhance predictions. The best model utilized Random Forest Regression, achieving a mean squared error of around 8 and an accuracy of over 96%.

Developed a model to predict sentiment (positive or negative) of tweets and reviews using Python's NLTK corpus dataset of tweets and IMDB reviews dataset. Implemented three methods: logistic regression, Naive Bayes classification, and word embedding, fully realizing the logic of all methods in models.