Projects

1. The system consisted of three microservices - front-end, order and catalog server with each server replicated and round robin is employed for load balancing.
2. The front-end server keeps an in-memory cache which stores recent results from read requests. Push based caching has been used wherein the catalog servers send an invalidate notification for the item to the cache before updating it in the database.
3. The application has each server on a docker container than can be easily be downloaded and deployed. For simple state level consensus, primary backup with fault recovery is used and Raft consensus protocol has been implemented for log level consensus.

1. Employed transfer learning using the BERT model and fine tuning it for downstream task of hate speech detection.
2. The model achieved 90 % accuracy on Twitter hatespeech test data.

Implemented DApp for Nim game server on Ethereum Solidity.

1. Implemented APIs for different functionalities like validating block and transactions, creating the merkle tree, keeping track of the UTXOs, getting tip of the blochain and so on.

1. Employed monitoring based approach by generating power curves using four parameter and five parameter logistic regression and mahalanobis distance.
2. Used 24 months training data from Bhopal Energy Center to generate power curve models.
3. The model gave a 95% accuracy on the test data.

1. Determined searchable interfaces using complex xpath queries.
2. Iteratively applied tf-idf against the content of the parent page and the results rendered from previous search to get the next set of relevant terms to be searched
3. Obtained good results on Indian Government research sites like icmr.nic.in and dst.gov.in.