Subratho Das

Innovative software engineer offering around 4 years of expertise in machine learning and deep learning flow. Seasoned professional with background in full data science lifecycle. Quickly learns and masters new technologies while working in both team and self-directed settings.

Applied machine learning : Applied AI

• Received Above and beyond certificate for successfully developing a Timeseries flow for AUTOML live project.

• “Fusion of ML models to identify sexual harrasement”, in ICCS Conference, Dec 2021

• Healthcare provider fraud detection : - https://medium.com/@subrathodas276/case-study-healthcare-provider-fraud-detection-d602e952ddac
• Sexual harrasment classification using ML : - https://medium.com/@subrathodas276/case-study-on-safecity-understanding-diverse-forms-of-sexual-harassment-personal-stories-4d31afcb213c

• Project: ATCC

Technologies/tools Used: Python, MosaicML, Mlflow, FastApi, Graphana, Prometheus

Description:

The solution is used to classify category and taxcodes (internal to company specifics requirements) for a given item title or details.

• Project: IDP:

Technologies/ tools Used: Python, ClearML serving.

Description:
The solution is an extension of ECM. All the newer integration of tax certificates will be done by the customer. This solution will significantly help reduce the efforts made by the AIML team each time a new certificate type is received for integration.

• Project: ECM

Technologies/ tools Used: Gitlab, Python, FastApi, ClearML, Crowdreason , LabelStudio, Scikit learn, Yolov5, Spacy, Helm, Kubernetes, docker.

Description:

A solution build to reduce the effort of an end customer, usually customer had to make manually classify region and extract information from tax documents. This system helps them classify tax regions and extract information from the tax documents.

• Project: Commodity Price Prediction

Technologies/ tools Used: Python, sklearn, streamlit.

Description:

A solution that will predict a commodity price based on time and area. E2e pipeline from training to predicting future commodities price. Showcasing the result using streamlit.

• Project: AUTOML (Data Science R and D team), Hypergiant – Live project

Technologies/ tools Used: Python, PyTorch, sklearn, SHAP

Description:

This is the research-based project to implement AutoML solution for data scientist, in this project we have implemented complete e2e pipeline for building highly efficient ML model, we have implemented highly concurrent solution which start from raw data and parallelly execute multiple pipelines with different types of algorithms for each stage like imputation, encoding etc. In this project we have implemented Efficient AutoML algorithm for Regression and classification using Deep Learning regularisers. Also, this Project has Univariate and Multivariate time series forecasting implementation. In terms of analysis this project has shap analysis integration for explainable model

• Project: Automated Data pipeline

Technologies/ tools Used: Python, sklearn, numpy, pandas, scipy.

Description:

A system that will create a pipeline for data preprocessing. Given data, users can customize the type of preprocessing to be applied. Generating a report which will have information related to the data and a data processing pipeline that can be integrated with Machine learning modelling flow.

• Project: MLops

Technologies/ tools Used: Python, machine learning models, flask_restplus, mlflow, lakefs, minio, airflow.

Description:

A system that will automate the training process. Data versioning will be done based on that training pipelines will get triggered. Flexibility in setting models hyper-parameter. It will help in managing multiple trained models through mlflow UI.