Responsibilities & achievements

*Drove the exploration of various neural network architectures for object and key-point detection using Python, TensorFlow, Keras, and the TF-Object Detection API. *Demonstrated advanced skills in data preprocessing and labeling. *Pioneered the implementation of an OpenCV algorithm to enhance hand image processing, further enhancing the app's functionality. *Orchestrated the successful deployment of sophisticated deep learning models on both iOS and Android platforms for seamless on-device inference. *Utilized Python to craft APIs, optimizing model deployment via Google Cloud micro-services and ensuring robust performance. *Marked a dynamic blend of research, coding, and cross-platform development, leading to impactful outcomes and enriched user experiences. *Responsible for spearheading the exploration, development, and seamless deployment of sophisticated deep learning and computer vision algorithms. *Leveraged expertise in deep learning techniques, and successfully enhanced object detection, recognition, and tracking functionalities, resulting in immersive and engaging interactions for users. *Worked on end-to-end creation of mobile applications for both Android and iOS platforms. *Demonstrated a profound understanding of deep learning, computer vision, Python programming, and mobile app development.

Responsibilities & achievements

*Conducted comprehensive research into state-of-the-art NLP techniques, employing them to create impactful vector embeddings and deploy effective ML algorithms for multiclass document classification. *Formulated and presented a compelling proof of concept proposal, showcasing a novel approach to custom document-type classification, contributing to client satisfaction.

Responsibilities & achievements

*Led a pivotal project for Legalario, involving the creation of an object detection API to accurately identify personal IDs from images. *Executed manual image labeling for personal IDs and skillfully trained an SSD neural network utilizing Python and the TF-Object Detection API. *Designed and developed a robust API using Python and Flask, capable of efficiently identifying, classifying, and processing valid IDs through advanced techniques in OpenCV. *Achieved seamless integration of the API into Legalario's systems, contributing to improved accuracy and efficiency in personal ID verification processes.

Responsibilities & achievements

*Spearheaded an innovative project aimed at predicting real age from single images, even without facial landmarks, showcasing a profound understanding of deep learning principles. *Employed Python and TensorFlow to develop a specialized variant of the Deep EXpectation (DEX) method tailored for mobile devices, demonstrating adaptability and efficiency. *Masterfully curated and augmented a diverse dataset of facial images, meticulously preparing it for training a lightweight yet highly accurate deep neural network. *Orchestrated the successful deployment of the model in an Android app, enabling users to capture and edit selfies for instant age prediction, showcasing advanced mobile app development skills.

Responsibilities & achievements

*Undertook a comprehensive initiative to classify dog breeds, culminating in the development of an Android and iOS app for dog breed identification. *Curated an extensive dataset comprising images of 139 distinct dog breeds, meticulously preparing it for training a deep neural network. *Exhibited exceptional data preprocessing and augmentation techniques, resulting in a model achieving an impressive accuracy rate surpassing 83%. *Crafted a user-friendly mobile application that enables users to effortlessly identify dog breeds by capturing new images or utilizing photos from their gallery.

Responsibilities & achievements

*Undertook a sophisticated data analysis endeavor, processing a labeled dataset of facial images to redefine beauty scores and classify them into distinct attractiveness classes. *Leveraged Keras and harnessed the power of the DenseNet121 architecture to construct a cutting-edge CNN model for estimating facial attractiveness. *Innovatively combined disparate datasets, encompassing human faces and random images, to train a model with the ability to discern facial images. *Demonstrated prowess in model optimization, converting models into efficient TFLite models and further refining their sizes using post-training float16 quantization. *Successfully deployed the models on an Android app, equipped with GPU delegates, empowering users to gauge the attractiveness of facial images with remarkable accuracy.