Mohit Raghuwanshi

User-Project Matching and Response Behavior Prediction using Autoencoder-Based Model. Developed a Siamese network inspired, content based recommendation model. Used one hot representations of attributes characterising users and projects. Used undercomplete encoder architecture for dimensionality reduction. Fine-tuned the model by adjusting hyperparameters and optimized the training process resulting in enhanced accuracy. Got an improvement of 15 % over an existing model. Tech used: numpy, pandas, python, mongoDB, matplotlib, neural networks, scikit-learn.

Developed a collaborative filtering model using the word2vec formulation. Constructed a corpus by extracting sentences from users’ search and view history. Utilized the embeddings in a recommendation system to improve recommendations. Augmented the model on a heuristic lookalike model in a water-fall fashion. Achieved a 22 % increase in recommendation metrics as a result of the augmentation. Tech used: python, numpy, gensim, scikit-learn, matplotib, sql.

Developed and implemented an optimized centralized architecture from a scratch with a common feature store for all the active/previous users. Facilitating platform by storing meta data, search, activity and response-based features of users to access them in real time or delay basis. Model development and experimentation cost reduced by large extent due to real time availability of features. Tech used: python,pandas, numpy, ML,Mongodb, RestApi, matplotib.

Performed Automation Manual Testing to create Gherkin Test Scenarios. Managed software development and delivered solutions based on Slice SLA assurance and 5G O-Ran. Tech Used : C++, SQL, jenkins, CI/CD, git.

Coursework included Regression, classification, SVM, KNN, Random forest, XGBoost, Deep Learning (RNN, LSTM, Transformers, BERT), NLP, TF-IDF, Word2vec, Data Science, Data Visualization, Data Preparation.