AI/ Deep Learning Engineer
Location : Noida, UP , India
Experience : 5 Year
Rate: $15 / Hourly
Availability : Immediate
Work From : Offsite
Category : Information Technology & Services
CHITRESH
DeepLearningEngineer
ADeeplearningengineerwith5yearsofexperienceindevelopingsolutionsinComputervision,andnaturallanguageprocessingdomain. As well as Hands-on knowledge in machine learning, deep learning, Tensor flow, Python, NLP, Understanding of functional design principles, object-oriented programming principles, basic algorithms
DeeplearningEngineer|21December,2018-Present.
Senior Deep learning Engineer.
ResearchandDevelopmentforDeeplearningandComputervisionsolutions.
Deploymentofdeeplearningsolutions.
Clientcommunicationsandprojectmanagement.
July,2016-June2018
CNNandTransformerbasedmodelwhichpredictsactivitieslikefight,fall,lying, Naturaleventslikefire,smoke.
It used 30 consecutive frames to analyze the activity and event in a video feed.Trained ona datasetcontaining approx.18000 samples, whileeach samplehaving 30images.
RealtimeinferenceusingTensorrt.
Back end prepared using flask app which can add and remove cameras dynamically andintegratedwithReact-app.
ProductdevelopedforRiyadhBasedclient.
Usedsweatporestoobtainfeaturematrixusingadistanceandorientationbasedfeaturevectorandobtainedsimilaritybetweentwofingerprintimages.
Usedsweatporestoobtainpatchesaroundit.TrainedaCNNbasedSiamesenetworkonthesepatchestoobtainsimilarity.
Useddeeplearningmodelslikekey pointdetection,objectdetectionandimageprocessingtechniques like perspective transform and Homographs matrix to map each player in avideofeedofmovingcamerasonstaticimageoffootballfield.
Usedobjectdetectiontodetectandclassifyeachpointinfootballgroundtogetcorrespondingpointsforstaticimageoffieldandframesofvideos.
Usedkey pointdetectionusingdeeplearningtogetco-ordinateofeachedge.
Usedcorrespondencetogethomographsmatrixandthenusedperspectivetransformtomap each player's co-ordinates on static field image, which were detected using againobjectdetection.
ObjectdetectionandOCRmethodtoextractnumberplatesfromrealtimevideos.Objectdetectionandtrackersfordetectingspeedofvehicleinrealtimefeeds.
InferenceusingTensorrt.
Back end and front end prepared using flask and react app.currentlydeployedonserversforFASTAGINDIA.
TMJanalysisbytrackingthepointonaclamptoanalyzethemovementofthejawofapatient.
ConvertedthePDF ofinvoicestoCSVusingobjectdetectionandOCR.
ItusedtodownloadthePDFattachedwithemailsandafterprocessingitsendstheCSVbacktothesameemail.
TrainedanOCRmodelontensor flowandconvertedtotfliteforandroidapplicationtodetectinformationwrittenonKenyannationalidcards
ObtainedaDistanceandnormalbasedfeaturevectorusingco-ordinatesandnormalofeachpoint,and thenobtainedoverallsimilarity.
Attendancesystemusingfacerecognition,inferenceusingtensorrt.
UsedSiamese,auto encodersforretrievingsimilarimages.
Usedauto encoderstodetecttheanomaliesusingunsupervisedlearning.
Firedetectionmodelpreparedusingobjectdetectionmethod.
Falldetectionusingobject,keypointdetectionandtransformermodelandtrackerstodetectfallofaspecificpersoninvideo.
Inthisprojectobjectdetectionmodelwasusedtodetectmen,women,andkids.Usingtrackerandobjectdetecttoanalyzethetotalnumberofvisitstoaparticularcounterofshop.
Backendandfrontendusingflaskandreactapp.DevelopedforRiyadhbasedclient.
Useddifferentregressionalgorithmsandensembleeachmodeltopredictfinaloutputofthemodel.
HighestKagglerank126.
UsedBERTmodeltotrain.
Useddifferentmachinelearningclassificationmodelandensembledthemtoobtainfinaloutput.
HighestKagglerank194
HighestKagglerank1201with99%accuracy.
Kagglerank308with97%accuracy.
UsedMLalgorithmsandBERTtotrainadatasetofmultilabels.
Predicteddifferentdiseaseslikepneumonia,covid19usingthedifferentconvolutionneuralnetworks.
TraineddifferenttypeofGANsandstudiedabouttheirapplications.
M.Tech - Communication and signal processingResearcharea:Computervisionandimageprocessing
B.Tech-ElectronicsandCommunication
Leadership skillsProjectmanagementCommunicationskills
MachineLearningImageprocessing
Natural languageprocessingDeeplearning
Computervision
PythonPytorch
Tensor flowandkerasTesnorrt
ONNX
Flask
I joined as a fresher in a startup in its initial days. I cracked a major project in 2month which awarded the startup its first major funding and kick started theirbusiness. It led to the recurring business for company and i was promoted as a ChiefTechnicalOfficerin11monthsofjoining.