Shubham Anuraj

@shubhamanuraj

  • I'm Shubham Anuraj.

  • I have made this blog from scratch.

  • The views expressed in the posts are purely mine or whoever the author of the post is.

  • This is also my portfolio page.

  • Projects

    • Generalized Game Playing Agent: A reinforcement learning (RL) agent with Monte Carlo Tree Search(MCTS) and deep learning to learn and play any kind of simple zero-sum game which has a terminal state pre-defined on CPU compute power. Neural networks help in better rollout estimation over time while the node selection is done through UCB1 formula.

    • AntiExcuseGPT: LangChain project built with OpenAI LLM APIs to generate reasons and inspiration to do a particular thing someone might be procrastinating on. Multiple responses chained together to give a comprehensive result.

    • Live Facial Recognition: Built an artificial neural network from scratch to classify family members who would be in front of the camera. Used the NumPy library to handle all the matrix calculation and OpenCV to extract RGB values of images from the webcam video stream and give a per second result of who is sitting in front of the camera.

    • Chess: A reinforcement learning agent which can play chess to some level of competency with just basic CPU compute power. Temporal Difference (TD) method used to update values of states along with a bit of alpha-beta pruning to reduce state space and get some decent estimate of the position.

    • Multiple Level Sentiment Analysis: A layered sentiment analysis NLP project which uses LSTM (Long Short Term Memory / RNN) which trains on reviews of hardware, to predict the sentiment and the category of the hardware being discussed. Full end-to-end pipeline created for classifying comments.

    • Character Level Language Model: 0.5M parameter language model from scratch trained on the book “David Copperfield” for generating text, given a prompt. This is a Transformer based model with learned embeddings.

    • Fine-Tuning Model for Sentiment Analysis: Using PEFT and LoRA for fine tuning a HuggingFace language model to complete sentiment classification for Yelp reviews. 88% accuracy after training.

  • Key Skills

    • Programming Languages:: Python, C/C++, Java, JavaScript, HTML, CSS, Bash, SQL

    • Applications: Large Scale Software Development, Machine Learning, Deep Learning, Neural Networks and Architectures, Mathematical Applications

    • Platforms: AWS, Docker, Kubernetes

  • Experience

    • Senior Software Engineer, Samsung Research Institute, Mar 2021 to Aug 2022

    • Software Engineer, Samsung Research Institute, July 2019 to Mar 2021

    • Intern, Samsung Research Institute, Jan 2019 to July 2019

    • Summer Intern, Nucleus Software, May 2017 to July 2017

  • Education

    • University of Birmingham, MSc. in Artificial Intelligence and Machine Learning, Grade: Distinction (81.2%), Northrop Grumman Scholarship

    • Manipal Institute of Technology, BTech. in Information Technology, Grade: 8.32/10

  • Resume for download.