Resume Akshay Kakkar

1. Akshay Kakkar
7028968747 | aksh.kakkar@gmail.com | Bangalore
PROFILE SUMMARY Skilled in Deep Learning, MachineLearning, Computer Vision and Python. Currently, working
on gathering data with exploratory and descriptiveanalysisof datasetand customizing,
trainingand integratingdeep learningmodels accordingto business requirement. Working in
the field of Artificial Intelligenceand MachineLearning for 3 years.
EDUCATION BACHELOR OF ENGINEERING IN Computer Science
Graphic Era University, Dehradun, June 2016
HIGHER SECONDARY SCHOOL
Army School Clementown , Dehradun, May 2012
SECONDARY SCHOOL
St. Mary Sec. School, Dehradun, May 2010
SKILL SET Machine Learning - Regression Techniques (Linear Regression,MultivariateRegression),
Classification Techniques (SupportVector Machines,NaiveBayes), Clustering(K-Means)
Deep Learning - Convolutional Neural Networks (CNNs), Fully-CNNs,Recurrent Neural
Networks (RNN) with LSTMs
ML Frameworks - Tensorflow, Keras
Programming Language - Python, C++, Java
Libraries - Numpy, Pandas,OpenCV, Matplotlib,Scikit-Learn
Software Tools – Jupyter Notebook, Anaconda
Web Technologies - Flask,Angular 6,HTML, CSS, JQuery
EXPERTISE Python
TensorFlow, Keras,
OpenCV, CNN, FCNN
PROFESSIONAL
EXPERIENCE
GRAMENER
Data Scientist - Feb 2019-Present
 Total Experience of 10 Months working on NLP and Computer Vision Projects.
 Currently Workingat Applied Materials ClientLocation and Involved in Computer
Vision Classification and ClusteringProjects for Semi-Conductor Images.
INFOSYS PVT. LTD.
Senior Systems Engineer- Jan 2016-Jan 2019
 Total Experience of 2.9 years working in Infosys R&D Department. Worked on
Various ImageAnalytics Projects which involved variousDeep LearningModels and
Computer Vision Techniques.
 Trained on Computer Vision & Deep Learning Techniques from Udacity after getting
selected from a programming competition held within company.
CERTIFICATIONS Trained by Udacity in “Self-Driving Car Nanodegree” Program

2. PROJECTS DEFCT CLUSTERING FOR SEM IMAGES
 Builta Pipelinethat would be ableto Cluster SEM (ScanningElectron Microscope)
Defect Images.
 Designed an Architecture that would take a set of defect images as inputand then
Cluster the defects based on the similarity between them.
 The architectureinvolved Feature Extraction usingResnet50 Pre-trained Model
(Transfer learning),Hierarchal Clustering,Outlier Removal and Cluster improvement
by identifyingWeak Clusters and Sub-clusteringthem.
DEFECT CLASSIFICATION FOR LASERTEC IMAGES
 Builta Pipelinethat would be ableto classify Lasertec Images (Semi-Conductor
Domain) between Pit and Particle
 The PipelineInvolved performing various imageenhancement likedenoisingan
image (Non-Local Mean), Image Normalisation,Feature Extraction (GLCM Features)
and Finally Classification UsingSVM.
SENTIMENT ANALYSIS ON CUSTOMER REVIEWS
 Worked on sentiment analysison customer reviews at Sentence level using
Language Model Like BERT and ELMO.
 Used various sentence breakingmethodology to get uniform polarity for one
sentence.
SEWAGE PIPE ANALYSIS SYSTEM
 Workingin a team for AnalyzingSewage Pipelines for different types of faultusing
MachineLearning/ Deep Learning and Computer Vision.
 Used Machine LearningTechnique Support Vector Machines to build a classifier
that detects faults and then fed the classifier to slidingwindowalgorithmto identify
faults.
 Developed Deep Learningmodel for pixel level classification of faultusinga Fully
Convolutional Neural Network (Semantic Segmentation).
 Used various Computer vision techniques likeMorphological operations,Sobel Edge
Detection for various purposes.
 Build the Entire system usingFlask and Angular 6.
RETAIL VISION
 Worked on Products Module in Retails AnalyticsProjectwhere we had to perform
various stock analysison the different products that are availableon a rack.
 Implemented various usecases for stock analysisusingtechniques likeYOLO for
Object detection and classification.
SEMANTIC SEGMENTATION, TRAFFIC SIGN CLASSIFIER & VEHICLE
TRACKING
 Findingthe drivableportion of the road usingFully-CNN. The solution involved a
pixel level classification of the drivableregion of the road.
 Identifyingthe various traffic signs usingLE-Net Architecture and then
verifyingiton the test set.
 Identifyingthe vehicles on the road by usingSVM, feature extraction and a
slidingwindowsearch to find the vehicles.
BEHAVIORAL CLONING
 Traininga car to drivea simulated environment usingdeep neural network. For
the data collection we drove the car in the simulator and then later augmented
the images obtained for the trainingdata for a more generalized result.