Dipak Kumar Tiwari
adngy9@r.postjobfree.com Mobile No: +918*********
https://github.com/tiwaridipak103
https://www.linkedin.com/in/dipak-kr-tiwari/
PROFESSIONAL SUMMARY
Working Professional with 2 year experience in the industry, with 1 year in data science and machine learning. Worked in
mechanical industry as graduate trainee engineer and after one year as Professional Mechanical engineer. Currently
completed training in data science from Applied Machine Learning Online Course.
SKILLS AND COMPETENCIES
Technical: Machine Learning, Deep Learning, Python, Natural Language Processing, SQL, Decision Tree, Random Forest,
SVM, GBDT, TensorFlow, Keras, Pandas, NumPy, Seaborn, Matplotlib, Clustering, Computer Vision. Domain: Machine
Technical, Product Management, Planning, Operational Research. Certifications/ Course Work: Applied AI.
PROFESSIONAL EXPERIENCE
Trainee Technician FUKOKU INDIA PVT. LTD. Apr 2019 Apr 2020
Work Description
Worked along with production and maintenance engineers and solves the daily problems with injection molding
machines.
Learned how to set daily target to sink up with future goal which required extensive dedication to complete the target.
Used to create a daily report and present it in PowerPoint to the senior executives after the completion of the project.
Used to inspect the product randomly in production line to make sure the quality of product.
Graduate Trainee Engineer Advance Cable Technologies (P) Limited Jan 2018 Jan 2019
Work Description
Attended daily morning meetings to set up the target for the day and performing all practical and administrative duties
assigned by the supervisor.
Traveled to other sites when need by the upper level and look after the assigned project and offer suggestions for
improvement.
Used to Present the report in PowerPoint and explained the ongoing progress of the project to the senior department.
SELF CASE STUDY
Instacart Market Basket Analysis (Predictive model)
Objective: Predict which previously purchased products will the user purchase in their next order.
Analysis: Identified useful pattern like there are some users who tends to shopping on every weekend for their weekly
expenses, Performed Univariate and Bivariate analysis on the data columns, Imputed missing values, tried using data
augmentation.
Models Built: Predicted the next items user would purchase using classification model. Tryout different classification
model like Logistic regression, SVM, Random Forest, GBDT and also MLP network .Used LightGBM for better
performance.
Results: Performance metric used was Mean F1 score. Above model LightGBM resulted in score of 0.38066 on private
score that would end up within 20% on Kaggle leader board.
Use-cases: The model outcome could help in more actionable operational changes and a better use of marketing
budgets for those companies who choose to use data analysis for better customer relationship by serving with right
product.
Blog Link :- https://medium.com/analytics-vidhya/analysis-of-instacart-from-kaggle-competition-12a26a24ef4b
Table detection and Tabular data extraction from Scanned Document Images (Table Predicting)
Objective: Detect the Tabular Structure if exists in the image and then extract the data from the tabular Format and
then convert them into a CSV File.
Models Built: Built TableNet using Pretrained VGG19 till Bottleneck as an encoder and two branch CNN network as
decoder model to generate table and column mask which helps in detecting tabular structure from given image input.
Results: Using TableNet model receive of 0.86 Recall, 0.95 Precision and 0.90 F1 score on train data and 0.61 Recall
, 0.90 Precision and 0.73 F1 score on test data . Increasing resolution size improves Recall and Precision.
Use-cases: The model can be helpful to detect tabular data from invoice pdf or image.
Blog Link : https://medium.com/analytics-vidhya/deep-learning-model-for-end-to-end-table-detection-and-tabular-
data-extraction-from-scanned-3eec3dce354c
PERSONAL PROJECTS
Scene Text Recognition Using ResNet and Transformer (Recognition model)
Objective: Predict the character string from input image.
Models Built: Built two models using the combination of ResNet and Transformer. 1 st model is the combination of
ResNet50 as an encoder and Transformer as decoder whereas 2nd model is the combination of ResNet101 for feature
mapping and input to transformer encoder and transformer decoder for predicting character string.
Metric: performance metric used was custom accuracy score which is the ratio of the total number of character
sequence matching for given predicted and ground-truth character string divided by the total number of characters in
ground-truth.
Results: Using 1st model receive of 0.87 accuracy and loss reduces to 0.0903 which is far better than 2nd model .Training
with more epoch would increase the performance of model.
Use-cases: The model can be helpful to predict word string from irregular and natural scene image.
Blog Link : https://tiwaridipak103.medium.com/scene-text-recognition-using-resnet-and-transformer-c1f2dd0e69ae
EDUCATION
Bangalore, India July 2013- July 2017
Bachelor of Engineering, Mechanical Engineering