Currently developing an augmented reality app to showcase items in an interactive way from our digital menu.

Implemented inventory system to reduce food wastage by forecasting future demand using LSTM and recommending a dynamic pricing strategy using raw material costs and their expiration dates with a Gradient Boosting Regressor.

We have impacted 9 clients across 3 states by decreasing their operating costs by 10%.

Link : MetaMenu.in

Built an unsupervised recommendation system that recommends a similar ticket to a given ticket. Reduced the time to resolve new tickets significantly. Exploited pre-trained word representations such as word2vec and GloVe.

Developed a Recommended Actions Framework that is used across the organisation in plugin mode and involved in full stack development as well as developing ML and NLP solutions for Enterprise use cases.

Researched and implemented a scalable clipboard management solution for the new Browser-based version of MATLAB

Handled various data types for clipboard actions with maximising support across multiple browsers.

Worked in startup at scaling stage, to handle webpage loading optimization. Decreased 25MB website to 8MB. Also learnt about search engine optimization.

Implemented elastic search in page catalog search engine, which improved the page searching capabiltiy, also improved performance and usability by adding fuzzy search concept.

Link : Green Farm Market

Lead an initiative with team of 2 students of my college.

Pool-in is a exchange platform for students to exchange the goods for low cost with other students. Like books , calculators and etc, within campus.

Derived an algorithm with a new loss function and gradients, optimised on hinge loss of pairwise distance between points which reduced the bias problem of Hierarchical Matrix Factorization algorithm.

Implemented the algorithm with metrics like ZOE, MAE and RMSE in Python, worked with MovieLens 100K Dataset.

Link : Thesis and Source Code

Used Neural Networks to automatically caption the images in Flickr 8K Dataset. Repurposed the InceptionV3 model by removing the softmax layer to enable transfer learning. This output of the network is fed as the image data to the neural Network. The Long Short Time Memory(LSTM) is used as the Network to process the partial captions generated after each iteration.

This is first cut model done as experimentation of combination of CV and NLP

Link : Source Code

Multiple face recognition in group photo, I used the MTCNN detector to extract faces from the given group image. Then trained faces extracted with Vgg-16 pre-trained model with modification of network.

Link : Source Code

Detection of Pneumonia in humans from chest x-ray scans using CNN. Used data augmentation to increase training data and model robustness to reach an accuracy of 90%

Link : Source Code

Developed an app which helps children with skill development in an interactive manner through Simon Says game.

Achieved Real-time pose detection with PoseNet and web app using ReactJs, won runner-up in Atmanirbhar Bharat Abhiyan.

Link : Source Code

Implementing a basic blockchain, proof-of-work, transactions, mining rewards & signing transactions using JavaScript

Link : Source Code