Part of the ML team at Apollo.io building next-gen GTM AI Assistant https://www.apollo.io/ai/assistant

Stybe is a one-stop-personal shop. The only constantly learning fashion app that combines AI-driven personal recommendations with taste-driven shopping.

Object Detection | Object Tracking | Domain Adaptation | Synthetic Data | Model Optimization

Building the next-gen AI-powered fitness platform. Worked mainly on 3-D Human pose estimation, activity recognition, and pose matching. Deployed solution both on the cloud as well as in the browser.

Provided consultation services to XTRA at a very early stage for setting up their computer vision and deep learning pipelines.

Working on addressing open problems in the domain of computer vision * Created module for automatic floor-plan (architectural drawing) understanding and recognition of objects of interest for DEWA (Dubai Electric and Water Authority), for fast-tracking the process of plan approvals. * Developed a COVID detection app in a great team effort, to identify COVID positive patients from cough audio recording. * Developed a system for denoising dirty/old documents than can be generalised to different types of documents. * Using a modification on CycleGAN for attaining the required results. * Also working on zero-shot super resolution, where super-resolved image can be generated for any given image with training on that image only, without the need of a higher resolution image.

Tech Stack: Android, Java, Python, C++, Linux, AWS, PyTorch, Scikit-learn, OpenCV, Flask. * Cheko is an IIT Kanpur spin-off startup in the anti-counterfeiting industry. * Worked on an Android app named 'Checko' using which you can scan and identify any tag as original (created by the company) or fake(created by the counterfeiter) within 3 seconds. * Worked on the complete pipeline, starting from data collection---data-per-processing-data cleaning--- modeling ---deployment. * My main work was to design a computer vision algorithm/pipelines so that any tag that is being scanned is identified correctly as per its true class with the least chances of error. * Worked on classical image processing techniques used for processing the input image; Have a good idea of keypoint detectors(SIFT, SURF, AKAZE, ORB, etc.), finding homography matrix between two images and other important concepts of image processing. * Worked extensively on OpenCV. * Designed a CNN from scratch to address the classification task, with size and time constraint on the model, as it had to be deployed on a mobile phone. * Deployed the model on mobile phone using PyTorch mobile, as well as Tensorflow lite--- To be decided later by company, what to release to the public.