Working on NER model incorporating Active learning to tackle low data resources helping MSME segments

Projects OverviewMedical Report Analysis: ► For helping the hospital chains examining their progress and making resources cost-effective► Designed a Medical Report Analyzer using bi-LSTM + CRF model; which extracts the entities related to drug, disease, body-organ from the patient digital report which in turn boosted their scalability in businessMemory Block Prediction: ► Catered the high memory latency issue of semiconductor firm► Developed a block2vec model to predict the next memory block to be accessed by the user given the previous block accesses using LSTM; reported 88% accuracy on ~0.2M data► Used in the cache optimizationProduct Details Recommendation: ► Co-worked with Computer Vision team to extract text and images from pdf and recommending similar products using Elasticsearch

Volume Prediction: ► To counter the problem of unutilized packaging space left on a delivery boy bag containing shipments and also to reduce human intervention► Built a Volume Prediction Model applying Doc2Vec embedding with RandomForest in Python; which predict the volume of a product using their description, the amount along with other features; resulted in the decline of median error rate from existing 48.4% to 14% on ~6M products► Later placed to production.Feedback Analyzer: ► Extracted the context and sentiment of the feedback received from the customers using LDA and word-embedding► Which helped to inspect the prospering and critical areas of the services

►Blur-Clear Images Classification: Designed a program to classify the blur and clear images; implemented through Neural Networks►Human Detector: Build a system to detect the presence of a human in an image, and achieved 97% accuracy on the home collected dataset of 834 images