S. Mohana Krishna

Master's student in computer science

About Me

Hi, my name is Mohana Krishna and I am currently studying master’s in computer science at University of Colorado Boulder. After graduating in Electronics Engineering from IIT(BHU)-Varanasi in 2015, I worked as a Jr.Research Scientist for one year and a software developer for 3 years at mykaarma.

As a junior research scientist I was involved in developing, training an NLP bot, whose objective was to suggest replies to incoming messages.We developed a game to crowdsource the task of generating training data and used generated data to train the bot. I was also involved in teaching other employees probability concepts as it was essential for developers to help improve their design process, conduct tests and grade them. This helped me improve my interpersonal skills.

Prior to that, as a software developer I was involved in various projects like designing APIs, implementing single sign using SAML, developing a windows app which could interact with browser using web sockets, develop a reporting dashboard showing different metrics to name a few.

My main interests are in the fields of Computer Vision, Machine leanring and Robotics.


University of Colorado Boulder

MS Computer Science

Aug 2019 - Present


BTech Electronics Engineering

JULY 2011 - MAY 2015


Human Interaction and Robotics Group(HIRO)


Graduate Student Reasearcher


  • Working on a project to improve existing object detection accuracy using boundary detection techniques.



Jr. Research Scientist

December 2018 - July 2019

  • Worked on the development of an NLP bot, whose objective was to be able suggest replies for incoming messages.
  • We had an internal game implemented to crowdsource the task of generating training data and used that data to train the bot.



Software Developer

June 2015 - Dec 2018

  • Was involved in developing APIs used internally as well as exposed to partners.
  • Was involved in developing a windows app, to support the ability to take retail payments from browser, which uses websockets to communicate with the app opened in the browser.
  • Was involved in implementing Single Sign On, using spring SAML Extension for the service provider and SimpleSamlphp for the identity provider, across mykaarma apps.
  • Lead the project of Internationalisation of myKaarma application, hence giving the ability to support multiple languages.
  • Developed and maintained a reporting application, which gives insight into the usage of the mykaarma application. Different metrics are calculated using mysql queries and resulting data is rendered as beautiful charts using google scripts.

WorldQuant LLC


Part Time Research Consulant

October 2014 - July 2015

  • Worked on creating alphas - mathematical predicitve models, to model the performance of financial instruments.

Karpa IT solutions (Filternet Foundation)


Software Engineer

March 2014 - May 2014

  • Made a basic working model of an image based adult content filter.
  • Approach was inspired from the research paper Detecting Pornographic Images by Localizing Skin ROIs’ by Sotiris Karavarsamis, et. al.
  • Using skin color information, identified regions of interest (ROI) from the image.
  • Extracted a set of 19 features from the ROI, trained a random forest comprising of 100 trees, with 3 features at most for a tree.
  • Classified images with an accuracy of 75%.


Autonomous Vehicle Competition


  • The core objective was to successfully navigate the car through a course track autonomously at a comparable speed.
  • Apart from the core objective, the vehicle was also required to perform the following challenges: backing out of a collision and performing a power slide during turns.
  • The navigation block of the vehicle was implemented using a state machine - which had inputs primarily from an RGBD Camera and sent appropriate commands to an ESC (Electronic Speed Controller). The ESC controls both the drive and steering motors.
  • The project allowed us to look at the real-world challenges of autonomous vehicles more closely, and provided us with a practical experience working with ROS.



  • Using basic image processing techniques like bilateral filtering, laplace transform and morphological operations have achieved an average dice coefficient of 0.862 on a subset of images from carvana dataset.
  • The time taken to process an image was 28s on average.

Semantic Segmentation


  • Implemented the following research paper DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs.
  • Used transfer learning i.e used a pre-trained VGG16 model, trained for the task of object detection with some modifications and re-tuned it for the task of semantic segmentation.
  • The model uses atrous convolutions instead of normal convolutions to tackle the issue of reduced spatial resolution due to repeated max pooling.
  • Augmented PASCAL VOC dataset of 8498 images was used to train the model.
  • The mean IOU achieved was 0.52 on validation set and 0.573 on test set.

Image Classification


  • Trained a deep neural network consisting of convolutional layers , max pool layers and fully connected layers on CIFAR-10 dataset to classify images.

Smart Cab


  • Applied reinforcement learning to build a simulated vehicle navigation agent.
  • This project involved modeling a complex control problem in terms of limited available inputs, and designing a scheme to automatically learn an optimal driving strategy based on rewards and penalties.

Adult Content Detection


  • Continued my work at the internship, mainly worked on increasing the skin classification accuracy of first stage.
  • Completed the project with a model using 19 features on a random forest comprising of 200 trees, with an accuracy of 80%.

A Little More About Me

My Skills

Programming Languages: Java, Python, C, MATLAB, SQL, Google Apps Script

Frameworks: Tensorflow, GWT, Hibernate, Spring

Web Technologies: HTML, CSS, CSS3, Javascript