- improved-ecomm | Customized ecommerce store application built with Django REST Framework, React, Redux and Docker
- Thumbnails API | Easy to manage, asynchronous images processing API built with Django REST, Redis, Celery, Postgres, Docker
- Library API | Streamline book reservations and manage library resources efficiently | Django REST, Flask, Redis, Celery, MySQL, Docker, Swagger
- Izanagi DAO | DAO Smart contract and fully functional front-end | Solidity, JavaScript, Next.js
- XAM | ERC-20 Token | Solidity, Python, Web3 py
- Demand Forecasting | Demand forecasting system for a short-term period (14 days) | XGBoost, ARIMA
- Brain MRI Segmentation | Brain tumor segmentation notebook using UNet as backbone | Tensorflow, Keras, CV2
- Domain Classifier | Classify malicious domains based on 3rd party data | XGBoost, Optuna
- Tabular Playground Series - Jan 2022 | Placed 53/1591 | Time series tabular dataset. The task was to predict a full year worth of sales for three items at two stores located in three different countries. | Used LGBMRegressor, XGBRegressor, CatBoostRegressor ensembled
- G-Research Crypto Forecasting | Forecast short term returns in 14 popular cryptocurrencies. | Experimented with a various models - XGBoost, LGBM
- Tabular Playground Series - Aug 2022 | Placed 874/1591 | Tabular dataset. The task was to predict product launch failures. | Used XGBoost, Shuffle & Split CV and Optuna for hyperparameters optimalization | Kaggle Notebook
- Tabular Playground Series - May 2022 | Placed 205/1151 | Tabular dataset. Binary classification problem that includes a number of different feature interactions | Achieved best results with DNN, PyTorch implementation | Kaggle Notebook
- UW-Madison GI Tract Image Segmentation | Task was to automatically segment the stomach and intestines on MRI scans. | Used pyTorch and 2.5D MONAI approach with UNet
- Google Universal Image Embedding | Task was to retrieve relevant database images to a given query image. | Used pyTorch, OpenAI clip, timm - blended timm with a clip model
- Jigsaw Rate Severity of Toxic Comments | Task was to score toxicity of a set of about fourteen thousand comments. | Used sklearn, nltk and ridge (a simple linear regression algorithm) to reduce overfitting
- Fake news detection | Explored a deep learning approach in fake news detection using BERT and RoBERTa | pyTorch, huggingface transformer | Kaggle Notebook
- Feedback Prize - English Language Learning | Task was to assess the language proficiency of 8th-12th grade English Language Learners