Data Science

Financial management firm client with core requirements to reduce the manual analysis of annual financial reports and get automated insights to predict future trends.
 

Implemented classical and advanced deep learning NLP techniques to provide an end-to-end data science solution from data collection, eda, preprocessing, modelling, evaluation, deployment, scheduling & monitoring.
 

Provided an API based solution to easily upload and analyze the documents automatically, resulting in the overall business sentiments distribution over the reports.

Client is a one of the major information security company looking for a solution to match unified customer identity across their large consumer base of ~12 million from different industries such as e-commerce, retail, digital services and CRM.
 

Built machine learning process pipeline to extract the data from different data sources, perform preprocessing on text, apply our matching mechanism & load processed output data into delta lakes.
 

We came up with a unified solution leveraging the use of big data, NLP and text mining techniques with PySpark on Databricks. 
 

Achieved a faster matching mechanism with a reduced processing time of 30% for low latency and quick output results.

Worked collaboratively with a major
asset manager to develop the process framework to automatically manage portfolios & predict stock performance.
 

Developed machine learning models to perform the predictions accurately with proper evaluation/validation strategies to improve the desired metrics. 
 

Built data pipeline to extract data, process and perform predictions as per scheduled timeline. Leveraged fully managed & monitored workflows using airflow and logging dashboards to keep track of visualizations and performance metrics.

Industrial manufacturer client with a usecase to automatically detect product grades from industrial machinery in real time.
 

Developed computer vision solution to perform classification tasks on different product grades on industrial machinery.
 

Used model compression techniques to deploy solution using IOT technology, resulting in reduced inference time to perform faster & low-latency predictions on micro-controllers.