Aayush Garg

Financial Marketplace for Indian Automotive Manufacturer:

• Led a team of data scientists to implement Machine Learning use cases for determining the chances of loan approval and prediction of Turn Around Time for loan application pertaining to purchase of automobile.
• The above use cases help customers identify the chances of approval of car loans from different banks/NBFCs and what will be the time taken by the banks for approval. Helps the customers make an informed decision regarding which bank/NBFC, they want to finance their loan from.
• Worked on exploratory data analysis, AWS architecture design, feature engineering, model development, hyperparameter tuning, and production implementation.
• The solution included using the historical loan applications, customer, asset and other datasets to create unique features and ML models.
• Delivered actionable insights from complex datasets through effective communication with key stakeholders across various departments.

Energy Price Simulation for Japanese Energy Trading Company:

• Led a team of data scientists to create a simulation API which provided client with an opportunity to plug and play various scenarios and keep track of revenues and other KPI metrics.
• Monte carlo simulation technique was implemented within the API so that the client could play with different values of selected parameters and select the best possible parameters to maximize the revenue.

Automation Project (internal):

• Worked on creating python automation scripts to reduce the manual effort for creating excel reports and dashboards.
• Automated the whole end to end process starting from reading the input data to creating excel reports using python and shell scripts with a single click.

Personalized Customer Engagement for US Retailer:

• Led a team of data scientists to develop customer engagement solutions.
• The solution included using the historical transaction, customer, product datasets to create unique features and rules that would result in targeting the most appropriate set of customers from the retailer's customer base and increase retailer sales and profit margins.
• The customers were given personalized offers (Rewarding Loyal Customers, Win back lapsed customers etc.) during a campaign based on their past engagement with the retailer.
• The impact of the campaign was then measured by comparing the uplift metrics of target vs control groups

Instore Customer Engagement for European Retailer:

• Developed and automated measurement solutions to measure the impact of an instore campaign. It involved developing codes and algorithms to analyse campaign data and extract metrics relevant to determining the campaign impact, reporting of metrics and impact.
• Developed a solution to determine the "LTV" generated from an instore media campaign using probabilistic model (Kaplan Meier Model). Using the model, was able to determine the % of customers that would remain engaged with a brand over a time period and the value contribution by them.
• Worked on creating a new measurement methodology where sales of a campaign were forecasted by using various time series forecasting and deep learning models. (LSTM, prophet, VAR, smoothing techniques, ARIMA models etc.). Able to achieve around 90% forecasting accuracy which helped in inventory management as well as measurement of impact of the campaign.

Anomaly detection:

• Project involved developing an Anomaly Detection Platform using Hadoop and Big Data Analytics to detect anomalous and fraudulent transactions across clients pertaining to different domains.
• Created analytical scenarios and rules to determine potentially fraudulent transactions using SQL and hive queries.
• Implemented machine learning algorithms, end to end automation using shell scripting and creation of interactive reports and dashboards using jasper soft tools.
• Feedback loop was implemented to incorporate the feedback on the fraudulent transactions from the audit team and improve the model accuracy.

PROCUREMENT

Worked to reduce the set of false positives using feature engineering and supervised Machine learning methods using a risk score mechanism

TRAVEL AND EXPENSE

Employee profiling to determine the outlier employees who are most probable to engage in fraudulent activities using clustering techniques for categorical and numerical data.

INSURANCE

Determine fraudulent transactions using supervised Machine learning techniques with data generated features.