I am an Artificial Intelligence enthusiast and I recently graduated with a BTech in Electronics and Electrical Engineering with a minor in Mathematics from IIT Guwahati. I am currently a Quant Research Analyst at JP Morgan Chase in the Market Risk group.
I am proficient at Python Programming (adept at Tensorflow, Keras, scikit-learn; working proficiency in PyTorch), C++ Object Oriented Programming and MATLAB Programming.
I want to perform state-of-the-art research in the field of AI and contribute in any way I can to the race towards making an Artificial General Intelligence. I am constantly aspiring to learn more from novel projects. My most recent work is on Reinforcement Learning for Games and Generative Models.
In my free time, I love to read non-fictional books. One of my favorite category is biographies of famous scientists. I am an ardent fan of biographical movies and documentaries. During my off time I love to travel in areas unexplored and 'untouched' by civilization. I am also a cricket fan and I love to play Chess.
Feel free to contact me for any opportunities related to Machine Learning or Artificial Intelligence.
I have been doing projects in AI both in internships and independently. I typically like to work alone but I have done some group projects as well:
I am always looking for projects with research aspects to them. I have been fortunate enough to have written three papers during my internships. They are listed below:
I try to write when I can. I write about anything that interests me which to be honest is an immensely broad category. I also am extremely fond of Urdu and Hindi poetry. I will try to link some of my favourite passages here too.
Check out some of my online articles here:
Developed a Deep Q-Learning Model along with a portfolio management environment for Reinforcement Learning. The model can learn to maximise either sharpe ratio or returns or minimize volatility. The model is able to beat the naive approach of assigning equal weights to all stocks when tested on several holdings of a popular mutual fund. Future work includes testing of various other RL models to improve the performance of the model and compare the performance with the performance of various mutual funds.
Developed a differentiating statistic between a spoof fingerprint and of that of a real person. Classification using this statistic was able to match the state of the art with over 90% accuracy. Various texture features were extracted and the feature extraction capabilities of Convolutional Neural Networks were also exploited to develop a robust fingerprinting recognition system against modern spoofing techniques.
On the recent conclusion of a course I took in the chaos theory, I came across the method of texture analysis using fractal dimension calculation of an image. Since fingerprints are visually self similar in the ridges pattern. I then used the box-cutting method for fractal dimension estimation and found that the dimension of the image is around 2.6 which is much higher than that of a non-fractal image's dimension which is usually equal to or below 2. Therefore I concluded that fingerprints indeed possess some fractal nature in them. Further work would be done to test this claim across datasets and applications of this discovery might be pondered upon.
Compared Vanilla GAN and DC-GAN for the task of generating face images from a small dataset. Researched on computational expense of GANs for small tasks and ways on how to improve it.
The task was to develop a sample efficient reinforcement learning (RL) algorithm to train an agent on the Minecraft game. Researched on various meta-RL methods for this sparse rewards and hierarchical nature of the tasks. Developed and tested Options-Critic Algorithm and compared its performance with several baselines like Rainbow and DDQN.
Constructed from scratch a bicycle which was able to balance itself and move along a given path without falling. The project’s motivation was the development of unmanned delivery systems for two-wheelers. Developed a novel technique of self-weight balancing by using PID Controllers to achieve self-stabilising action in the bicycle.
The task was to generate similarity scores between a customer query of a job and hundreds of options of Resumes for finding the perfect match for the job. Challenges included short text nature of the query and scalability. Developed a software which uses word embeddings from pretrained word2vec algorithms and various similarity metrics for each option to find the best match. Keyword extraction performed over Resumes text and scraped data from LinkedIn profiles.
Used image processing techniques over images from a blood smear on a glass frame for detection of diseases like Malaria. The algorithm was able to report RBC, WBC and Platelet counts and thereafter reported a probability of being ailed by any of the five diseases we trained the model to detect.
Current CGPA:
Major: 8.76/10
Minor: 8.63/10
Class XII - 93%
Class X - 10.0/10.0 GPA
