Building AI products with user trust as the main axis

Surendranath Jillella  • Peerlist

- As a company, we are looking at developing solutions using Xpresso for various clients and deploying  them on premise. This comes with a challenge of integrating the solution with the client environment which takes effort and time away from development and also gives new complications to work on. - The aim of this project was to reduce this effort on environment setup and make it seamless to deploy the core parts of the solution without any hassle. - Work involved containerizing Xpresso using Docker. And automating the deployment process beginning from building docker images, hosting the docker images and version control using DockerHub.

Containerization of Xpresso using Docker

- This project aims at classifying emails based on the content of the mail into 10 different categories. - This was a solution to a retail giant to help their HelpDesk team process the merchant and supplier emails efficiently. - Worked on various Natural Language Processing techniques like dependency parsing, POS tagging, Lemmatization and Named-Entity Recognition for retrieval of semantic information from the text of email. - Also responsible for deployment of solution in client environment. - Also part of the client communication team.

Email Classification

- As a customer, we are frustrated when we don't get food we ordered. From a restaurant manager's perspective, this doesn't reflect well on the restaurant and is expressed through various channels of social media. This could degrade the restaurants' image effecting its sales from new customers as well as repeat customers. - The aim of this project was to identify such accuracy issues in food orders for a restaurant-chain company from customer reviews sourced from various channels like Facebook, Twitter, Zomato, Email Feedback. - Work involved developing a rule-based NLP engine to parse the customer reviews and identify if there was an issue of wrong order being placed or delivered. - This has helped the management to identify the problem areas and enhance their customer experience.

Identification of Accuracy issues in Food Ord

- Aim was to develop and NLP engine to process the adjuster notes to extract semantic features for Machine Learning model to find the probability of Subrogation in an insurance case. - Exploration was done on the data provided by client and features identified. - Implementation of an NLP rule-based feature extraction module. This helped us identify the semantic information from the note instead of just working using a keyword based system. - Work involved NLP techniques like dependency parsing, POS tagging, Lemmatization, True Casing and also word vectors to extract features. - This project also involved an Machine Learning module developed in Python to give the prediction based on the features extracted from NLP engine.

Inference Engine to predict Subrogation

- This project aims at monitoring information sourced from various news websites, RSS feeds, social media posts, etc. to sense the sentiment of the citizens on various aspects and find the topics trending  in a particular geographical region based on the frequency and recency of their mentions in multiple data sources as mentioned above. - This is a client solution based on Xpresso. - Work involved aspect-based sentiment analysis, trending topics detection, customizing Xpresso to the needs of the client. - We are also working on a user feedback mechanism through which the user can monitor sentiments on new aspects(as the importance of aspects keep changing with time, causing emergence of new aspects to monitor) apart from the predefined generic aspects.

Sensing Engine

Abzooba's core product Xpresso is a knowledge distillation engine that processes unstructured textual information fetched from online review sites, social media, blogs etc. and extracts knowledge relevant for decision making. Aspect-based sentiment analysis, statement type determination, emotion detection are some of the key capabilities of the engine.