BYJU'S App Profiling and Memory Leak Identification on Android and iPads Senior Research Engineer Employment
Aug 2020 - Apr 2021
- Memory profiling of iPads using XCode and Lenovo Tablets using Android Studio
- Identified a possible kernel bug in Lenovo 605LC variant that was causing crash in 1000s of customers of BYJU'S
- Working with Lenovo Team to help them reproduce the issue and solve the memory leak of the specific device variant - Lenovo 605LC
Skills used: XCode, Android Studio
BYJU'S Production Pipeline Management Senior Research Engineer Employment
Mar 2020 - Apr 2021
- Worked with the Product Management team of Worksheets team as a Point of Contact for Vision Team for all CV and Machine Learning related workload division
- Involved in review of the books for PreK to Grade 3 kids of various subjects including English, Maths and Science for India and USA region for CV compliance
- Helped Setup the pipeline with the Project manager for the Quality Assurance QA team, the Dev team, The UX team, the CV team to get better throughput and minimise turn around time.
- Helped in development of tools (online and offline) to help people with the productivity for better efficiency in the pipeline
Skills used: Quality Assurance QA, Product Management
BYJU'S CV based Line tracing over sheets drawn by kids and Auto-tagging of reference images Research Engineer Employment
Oct 2019 - Mar 2020
- Computer Vision was used to detect lines drawn by kids for Tracing game and filter out if they are drawing gibberish
- Automated the tagging tool for the tracing game in the Dart Frontend which reduced the time of Tagging Team from 15 min to a single click.
- Graph theory was used to figure out the strokes in sheets which contains letters, numbers and images in dot tracing format
- Image processing was used to detect the dots on the canvas from the reference image for auto-tagging
- Made relevant changes to the back-end in Go to accommodate the requirements on the Data Store in Google Cloud.
Skills used: Computer vision, Dart, Go, Google Cloud
Whodat Monocular depth estimation and 3-D reconstruction Deep Learning Engineer Employment
Oct 2018 - Jan 2019
- Deep learning models such as DeMoN and MVDepthNet were used to explore depth using single camera in Python and Tensorflow
- Computer vision was used to stitch together the depths to generate the 3D scene
- Computer Vision was also used in figuring out the poses between recurring images using visual and dense features
Skills used: Computer vision, Deep learning, Python, Tensorflow
IIT (BHU), Varanasi Network Security Lab Resource Management Student Passion Project
Dec 2016 - Jun 2018
- Setup a cluster of 10 PCs to speed up the computation for Tensorflow using Hadoop
- Worked with CLIs and bash scripts to manage the system remotely. Used various port forwarding techniques to ensure proxy free internet for the devices to get proper updates
- Setup a GPU Server and helped with the load balancing of the users based on resource utilization for uniform resource division.
Skills used: Tensorflow, Hadoop
UC Berkeley Neural Programmer Interpreter Student Internship
Jun 2017 - Aug 2017
- Mujoco simulator was used to simulate a fetch robot and generate data as we didn't have access to the actual robotics system
- A paradigm of Deep Learning, called Neural Programmer Interpreter, was used to learn from the data generated and generalise to unseen tasks, with code being written in Tensorflow
- Designed the stacking problem so as to generalize across a varying initial state, configuration and across different number of blocks
Skills used: Deep learning, Robotics, Tensorflow
IIT (BHU), Varanasi One-Shot Learning using Memory Augmented Neural Networks Student Passion Project
Dec 2016 - Jan 2017
- Experimented few-shots classification using a paradigm of Deep Learning called Memory-Augmented Neural Networks on Omniglot dataset.
- Implemented novel Memory-Augmented Neural Network library using Python and Tensorflow library and published on GitHub.
Skills used: Deep learning, Python, Tensorflow
Link to the github: https://github.com/hmishra2250/NTM-One-Shot-TF
IIT (BHU), Varanasi Botnet Detection in Computer Network using Machine Learning Student Course Project
Jul 2016 - Nov 2016
- Implemented a novel scaleable flow generator based above SFrames, which could easily scale beyond RAM as opposed to Pandas DataFrames.
- Extracted useful Network flow based features from the raw Packet capture files and also based on network-flow domain knowledge
- Trained simple models: Classical Machine Learning models, Ensembles and Simple Feed-Forward Network and reported thebest model performing best on test set
- Detected the botnets in the network using the models and features computed for security of the network
Skills used: Security, Machine learning
Link to the github: https://github.com/hmishra2250/Botnet-Detection-using-Machine-Learning