Abhishek Malik

Generated artificial banking datasets and trained machine learning models using XGBoost to demonstrate our proof of concept to potential clients. This proved crucial in acquiring first customers (banks).

Over a year experience of building and deploying models into the Hawk:AI Java based platform.

Made the models explainable using SHAP values, so its predictions could be interpreted by any bank operator.

Trained an LSTM-VAE (Variational Autoencoder) using account balance data (time series) and detected outlier accounts. A quick demo and source code can be found on this link.

Extracted 300k time-series features using TA-Lib, trained a VAE and identified suspicious transactions.

Manually engineered 80 behavioral features – which expresses changes in usual account activity.

Trained a LightGBM model on this data to explain predictions of a Deep Learning model (Autoencoder) which generates such features automatically. These models were used to identify suspicious cases.

Built a tool that produces textual explanation for every prediction & helps investigate identified suspicious cases.

A demo of the tool can be found on this link.

The program systematically edits the HTML code on their Moodle website which was only done manually until now.

The tool is written in Python. It uses APIs & MySQL to access database and uses Django for frontend.

Developed a novel machine learning method that outperformed current conventional planet detection methods.

Method used TSFresh for feature generation and XGBoost. Built a vetting-tool which let people explore predicted planet candidates from Kepler data, a demo can be found on this link.

Presented a poster at 711th WE-Heraeus-Seminar. A publication on the work is currently in progress.

Wrote python scripts which enabled a 90% increase in computational efficiency of the program by eliminating GUI

Conducted simulations using High Performance Computing (HPC) facility & studied the phenomenon of drag crisis as a part bachelor project.