Having a curious mind and always willing to learn, I am an enthusiastic Data Scientist & Software Engineer who is ready to take challenges and provide something state of the art not only to the Science community but also to the Business domain. I work primarily as a Software Engineer at Scrub AI Holdings Limited in the UK. On the other hand, I love doing reseach in the domain of Computer Vision and Human-Computer-Interaction. However, scrutinizing the recent technologies regarding Blockchain and Quantum Computing is another realm of my passion. I love traveling and reading books while doing voluntary research in Deep Learning and AI. Anyone willing to collaborate/ Learn and Teach in any of the domains of my interests is always welcome.
Completed the most exciting and important phase of my student life where I got to collaborate in many research projects not only confining myself within the boundaries of the university but also outside the frontier of the country. My undergraduate thesis project was Blockchain Based Categorical Information Transaction and Depository Framework.
Mentoring and Contributing to an Open-Source project that aims for AI4GOOD. Collecting data from different online and news sources along with the help of Sigmind.AI, who provided with road-vehicle image dataset built by them, the project was a brilliant contribution collaborating with people from all over the world.
View Project
In this project my primary goal was to build a Unpaired image to image translation using Cycle-Consistency. Such such architecture I had to implement Cycle Consistency, Least Square and Identity Loss while building a U-NET architecture. This work was inspired from this paper and with the help of DeepLearning.ai
View Project
Utilizing conditional GAN I implemented a Pix2Pix GAN that can convert aerial satellite imagery ("input") into map routes ("output"), as was done in the original paper. The architecture for the generator is a U-Net and is implemented from this paper and with the help of DeepLearning.ai
View Project
Fine-Tuning Huggingface's Pegasus Transformer model using Google's PAWS along with MSR Paraphrase Corpus dataset, I built a paraphrase machine which is able to rephrase any english sentence of short and medium length. I used Django Framework to build to end user system.
View Project
In this project my primary task was to generate a dialogue between two bots by feeding conversations to the Reformer model and thus it will learn how to understand the context of each one. Not only will it learn how to answer questions but it will also know how to ask questions if it needs more info. I preprocessed the MultiWoz dataset for Reformer inputs and fed it into the pre-trained Reformer Model
View Project
Training an DJ Model (LSTM Network) on a given dataset provided bt DeepLearning.ai. The task primarily involved preparing the training inputs and feeding it into the LSTM model. There were 78 unique values(notes) used to train and build the model.
View Project
Implemented a simple Auto-Correct algorithm using bayesian probabilistic model. For this i had to apply Minimum edit distance while switching different letters for auto-corrention as well as word recommendation.
View Project
Bi-Directional Encoder Representation Transformer Abbreviated as BERT is an Encoder Transformer Networks proposed by Google in 2018. In this Project my primary objective was to use a pre-trained BERT model and preprocess the necessary data (such as masking and pairing) to feed into the model and finally evaluatng the models.
View Project
In this project my task was to summerize news articles using self made transformers model. The dataset used was from cnn_dailymail. Implementing both dot product attention and casual attention I built a complete langiage model which was used to summerize the texts given as input. Finally using greedy decoding as evaluation metrics I evaluated the model's performance.
View Project
In this project, my primary task was to construct a speech dataset and implement an algorithm(GRU) for trigger word detection (sometimes also called keyword detection, or wakeword detection). My trigger word was "Activate." Every time it hears me saying "activate," it will make a "chiming" sound.
View Project
Building a Neural Machine Translation (NMT) model to translate English-to-German texts. To achive this task, an attention mechanism was used to allow the decoder to access all relevant parts of the input sentence regardless of its length which can solve vanishing gradients for long sequences in the RNN along with LSTM network.
View Project
Built the YOLO Object Detection algorithm by filtering with a threshold on class scores and Non-max suppression. Finally I evaluated the model using box-confidence. This project was an Assignment from DeepLearning Specialization by DeepLearning.ai
View Project
