The end users in the organization found it difficult & time consuming to raise an IT service ticket on the IT Service request tool. The IT service provider was hence looking for an innovative solution which will be easy for the users and improve user satisfaction.
Based on the business problem, we advised to have a Telegram APP where the users can quickly report the IT issue by simply mentioning about the issue.
We then developed an NLP, NLU, NLG & Deep Learning based Engine (BOT) to extract the user messages from the Telegram App, respond to user queries, create tickets in the ticketing tool (ManageEngine), update users about the ticket status.
The NLP Engine (BOT) was developed using the Custom AI Algorithms.
The customer is working with state level security agencies to provide intelligence from crime data. They were looking for social media data based insights for the habitual offenders or people with suspicious activity.
Python, Django, NLP using spaCY
The customer was a state level security agencies and already have a solution to register Crime data & criminal database. As an extension to current solution, the customer wanted to explore use of social media analytics to provide additional insights based on sentiment and network graph based user information. The solution is developed for Twitter data source. The solution could give 4 level deep twitter id connection, but also provided sentiment analysis of tweets posted by identified twitter handle. This was a PoC, and it is still under discussion.
The customer wanted to automate the note taking process of executive conference calls and sales review calls internal to their company. The current process was manual, which was error prone and time consuming.
Python, Django, Google Speech to Text API
The speech to text solution was developed using Python & Google speech to text API. The application frontend was developed using Python Django framework. These speech to text API provided good accuracy (~90%) for native English speaker and when number of speaker are less than 3. The accuracy reduced when multiple speakers with vernacular accent were speaking. Also the current version of Google API is not matured for speaker diarazation. Ran the pilot project and demonstrated results to the senior stakeholders
The solution is implemented for customer service desk for B2B operations of leading telecom provide in India. They were monitoring 8 mail boxes related to Billing, Customer enquiry, Activation, Termination etc. It was a manual process and many times the emails were not replied on time or missed to reply. Also average handling time was significantly higher
The solution is implemented on premise at customer site. The solution is customized to track specific KPI’s the customer wants to measure. There were total 8 categories in the scope. Initial version AI models were trained on 8 categories. These models were retrained in ongoing basis after 1 month for the period of initial 3 months. Entire solution was implemented in 6 weeks duration,
The KPO was looking for innovative solution to resolve any Insurance Claims processing related queries which the Agents might have while they on the claims processing. The KPO also wanted a mechanism where the agents are made aware about the claims processing policy changes made by the end customers, as and when the changes are broadcasted by the end customer.
As per the business problem, we created a conversational BOT as a web application. The BOT was trained on the Knowledge Base related to Insurance Claims processes. The agents could choose the topics on which they have query and ask for the information. The BOT could answer the queries, ask feedback related to the answers, suggest potential questions which the user can choose from.
If the user still did not get the right answer, the BOT could involve the supervisor in the conversation. The BOT could use such conversation for re-training.The BOT was also used to broadcast new policy changes and training details to the agents. The BOT has now become an integral part of the KPO Ops team
Information extraction from Salary slip and Employment Offer letter is key requirement to understand income of the mortgage applicant. Another challenge is the document structure is not fixed for these document.
Python, Django, NLP using spaCY , BERT
The requirement was to extract Employee Name, Company Name, Date of Joining and Compensation from Employment offer letter. Used Named Entity recognition based solution to extract these information from the document.The salary slip need to parsed to extract key information like Employee Name, Employee Number and Salary details. Used combination of NLP techniques like BI-LSTM and Computer vision based table data extraction to extract required this information
The smart city solution focussed company wanted to automate routing of customer complaints by accurate classification. The current manual categorization of customer complaints is error prone, results in rework and takes This will result in faster turnaround time for the customer complaints and improve citizen experience
Python, Django, NLP using spaCY , BERT, XGBoost
The complaint need to be classified into 150 categories which consist of up to 4 level of depth. Implemented classification with XGBoost based approach initially, but accuracy was not as expected. Implemented category classification using BERT which achieved validation accuracy of more than 90%. The training data was inadequate, this data was augmented using NLP augmentations techniques. Built the model retraining pipeline and automated the model training process.
Our client was looking for a solution to automate the Quality Assurance process of analysing the Collection process call and build analytics. This would be used to evaluate the Agent and Customer on multiple parameters to have objective analysis of the call at scale.
Python, Django, React, BERT, spaCY
AI based Analysis of agent and customer for the collection process help to understand Agent performance and Customer Behavior.
Using the audio recording transcripts were generated using which various analysis tasks were performed. Relevant information about each agent and customer were parsed and stored in the database. Using this information, various KPI’s are derived, which is used to identify agent training need and adherence to compliance requirement
Detecting duplicate claims is a major challenge facing to Insurance industry. Also there are instances of document or image tempering
It is easier for anyone with basic knowledge of Photoshop to falsify digital images or scanned copies of official documents from birth certificates to bank statements
Python, Deep Learning
Image Tempering : Our advanced AI based solution detects various types of image tampering like
Our solution classifies the document as tampered or not with the confidence score. AI based solution learns continuously to improve accuracy
Every month thousands of invoices data entry is done by the BPO company. The client wanted to automate manual data entry process. As a part of the process, product category for grocery or food line items need to be looked up manually and then enter it into the software. This require a significant work force to perform this activity
Python, NLP , Deep Learning, OCR
The solution involved extracting data from invoices using our document data extraction tool. We have developed end to end flow in our solution, which involved picking Invoice data from shared folder, extracting data and integrating it with their downstream system. Once the data is extracted, the line item description was passed to Invoice Line item classification API to classify it in right category.
The training dataset provided consisted of vendor, item description and category of the food item.
Currently the client has team of agents, which call candidates and employees to conduct the survey. They were looking at automating this process with the help of chat bot.
Python, Chat Bot
We have implemented out chat bot platform for this requirement. The main components of this platform are Interactive AI based chat bot, Actionable Analytics and Admin functionality to manage the bot. The client wanted to customize this solution to add reward management functionality and some custom analytics based on survey response. The client was able to scale up non linearly to serve more customers, than current model. We delivered the entire platform in 8 weeks time including customization
A leading educational institute wanted to implement a solution, which will analyse research article & generate a short summary of it. The short summary will be rephrased to avoid any plagiarism charges. Post human verification, this short article will be posted to social media sites like LinkedIn.
Python, NLP, Deep Learning, LinkedIn APIs
The client wanted to save efforts to publish blogs and looking at automating this process using AI. We have developed a Natural Language Processing based solution, which understands the context of the document. Post that, there is an option to generate percentage summary of the whole document. If we choose 25% summary, then main document would be summarized to 25% of the main content. Post that, we have used a third party to API to re-phrase the content. A web application is developed to view results of original & re-phrased summary. The solution provides capability to verify and correct the summary. Post this, the article will be published automatically to LinkedIn
The client provide service to review contracts and share findings/recommendations to their client. The current process is manual, where team of lawyers go through each contract and extract facts, obligations & risks. They share this report over the mail to their client. This is time consuming and error prone process
Python, React, NLP, Deep Learning
We have developed the contract review platform, which is deployed as a SaaS solution. This platform enables the client to upload the contract. The AI based document data extraction engine parse the contract and extract Fact, Obligation and Risks from the contract. The AI model was trained on more than 5000 documents. The human in the loop approach provided in solution enables the lawyers to verify the extracted data. The solution provide mechanism to collaborate between various stakeholders. The reviewed contracts are then share with the client on platform itself.