My Portfolio

Prabakaran Chandran B.E., Data Scientist

Helping Fortune 500 Firms with AI and Data Science

Data Scientist with 2+ years experience , have been working with fortune 500 clients to solve their problems using the power of artificial intelligence and data science

As an Avid learner , I am always looking for learning and experimentation opportunities and passionate about upskilling myself and others

As a part of my Data Science Journey I have been mentoring colleges and students and enabling them with Data science and machine learning skills

Work Experience and Projects

Data Scientist – Mu Sigma Inc – 2019 June – Present

1. End to End machine learning pipeline to predict the budget for project portfolio

Have built an end to end machine learning pipeline to allocate the monthly budget for each construction project portfolio for a Japanese construction firm. Trained an ensemble and deep learning based models using building properties and monthly budget distribution. The model was deployed using a flask application in the user environment.

Tech stack used : Python , scikit learn , keras , flask , pandas

2. Data driven failure management system development using r shiny

R shiny based Web application for a Japanese tech firm to understand about their server failures, shipments, and Repairs was built in order to carry out prescriptive maintenance of servers. Application has served important KPI metrics to track the operations around shipments and returns

Tech stack used : R – DS packages and R shiny

3. Agent based modeling to predict the incidence of beverage drinkers

Bayesian network based agent based modeling framework to predict the consumption of soft drinks in USA. This Bayesian network was built based on consumer behavior , their preference and impact of marketing events. The Conditional probability table was used in the real time simulation of agents to determine the amount of incidence every month. The framework has helped the US beverage giant to consumption behavior and impact of the firm’s activity on the incidence

Tech stack used : R – DS Packages and R shiny

4. Market share analysis of Soft drinks

A Penalized Regression powered market scenario builder was developed to simulated the state of market shares when the price of competitors’ soft drinks and firm’s drinks change. Regression algorithms used to capture the relationship between price and market shares. Final Regression equation based on elastic net model was used as a base function of the market scenario simulator. This prescriptive analysis helped the Canadian beverages firm to play around the different price ranges and foresee the impact on market shares.

Tech stack used : R – DS packages , Excel and SQL

5.Computer vision based Defect detection system for Modern manufacturing firm

A Deep learning based Defect detection model developed using Faster RCNN with ResNet 50 DC5 backbone. Before to that various backbones were trained and validated. The validated Faster RCNN – Defect detection model has 99% of recall and 90% of mAP value.

In order to filter out the defects occurred in the product area , a Mask RCNN – ResNet 150 based Instance segmentation model was developed , later the Product area segmentation model was integrated with Defect detection model. The final integrated system is able to reduce the quality assessment hours from days to hours

A 3 tier architecture based Application was developed to use the integrated deep learning model in the real time. Django based Rest API is act as interface layer between printed product images and the integrated model. A multipurpose user interface is developed using React components in order to provide the detailed defect level information and insights around defect occurrence.

Tech stack used : pyTorch , Detectron2 , tensor flow , Django ,React , Postgres , Django Rest , Material UI

5. Genetical Entity Recognition (Named Entity Recognition) using Transformer models

Bert and GPT based Named entity recognition models were developed to extract the genetical entities from the bio medical research papers. The final model deployed using fastapi. The Biological Bert model was the best one since the model was already pretrained on the huge medical corpora. The GER application helped the research team to filter the research papers on genetics and search for specific enzymes and protein related papers

Tech stack used : Hugging face transformers , pytorch , fast api