Project: Transaction Classification System- Project Goal: Built a Classification System as POC to classify customer transaction data into different classes and built customer spending profiles and save INR 60-70 Lakh annually in cost by replacing third-party paid services.- Conducted data cleaning using Pandas, NumPy, NLTK, Databricks and Converted transaction narrations into vectors from word embedding created using Word2Vec and DistilBert Tokenizer and applied various Clustering ML models to create meaningful clusters of the transaction narrations.- Leveraged Machine Learning and Deep Learning based models to classify the previously clustered narrations into predefined classes with varying levels of accuracy based on the model used.- Optimised DeepLearning-based LLM model by converting to a Quantized ONNX model and achieving an accuracy of 96% and increasing inference speed by 8x-9x times on a CPU-based machine hence reducing execution cost.

Project: Store’s Sales analysis and Future sales predictions.- Project Goal: Analyse historical sales data of different stores and predict future sales.- Performed Exploratory Data Analysis through Data Pre-processing and Data Visualization to deliver data-driven insights to the stakeholder and further utilised the insights in building the Prediction Model for future sales.- Analysed and filtered 850k rows of data of 150 stores and approx 1800 items received in multiple batches.- Built and tried different Machine learning models to predict the future sales of different products in different stores using Scikit-Learn and Optimised models by tweaking the Hyperparameters of the models to achieve 88% accuracy.