An enterprise cognitive search engine based on the IBM Watson platform, that takes a user query in Natural Language and find an exact answer from Structured (CSV, Excel, DB, etc) and Image files. Hosted it on the cloud using Amazon Web Services.

Generated semantic graphs for organization policies so that users can easily get to exact answers rather than going through the whole policy document. Sentences of policy documents were tokenized, lemmatized using spacy, wrote grammar rules to fetch out triples from the sentence. In the last stage, these triplets were used to generate graphs using the Cayley Graph database.

To generate the dataset for bot frameworks by fetching intents and questions from chats/Emails. An unsupervised learning problem was converted into a supervised problem using K-Means and TF-IDF and then using supervised learning classifications algorithm (RF) I got desired results (questions and corresponding Intents). Used Feedback mechanism to improve accuracy with time. Hosted on IBM Cloud using Python and Flask.

Built an algorithm to generate a model for Revenue prediction for an organization at various levels (Business Verticals and Account Level (>1000)) also generated Inferences. For prediction, I used LSTMs (Keras) and other python libraries. For Inferences, I used Bayesian Networks (bnlearn - R) and tried Bayesian Neural Networks (Edward with TensorFlow). On this, I achieved an accuracy of >98% at the Vertical level and >90% at the Account level.

Built multiple backend services for the website using GoLang. Enabled full text search on CouchDB using Lucene Search and exposed rest endpoint using NodeJS. Created excel parser using python to parse financial sheets.

Built a recommendation system for an e-commerce website. I also wrote an algorithm for pricing the products for that we considered multiple factors like the shortest route, time to deliver, delivery partner, etc.

Created multiple backend services for fleet management service. Enabled full-text search using Elastic search. Included caching for faster access to the database.

Created a dashboard for ERP solutions at AOB. Analysis of multiple business matrices for available inventories, production analysis, finished goods inventories, revenue reports, etc. For the pipeline setup, we used Kinesis (to capture the events), S3(to store the events). Transformed those events for further use. Used AWS QuickSight for dashboard and reporting.

Implemented a domain-specific Knowledge Graph (KG) from legal contracts that can be queried to answer for contract analysis. Custom NER model for the legal domain to extract key information from the documents. 
Used: Python, BERT, SVM