Artificial Intelligence

Research and Conceptualization

Research and Conceptualization

All the required research including assessing products already exist in the market and run a gap-analysis.

Testing the Idea for Feasibility

Testing the Idea for Feasibility

We guide you beyond just product conceptualization through proof of concept (POC).

Architecture of Product

Architecture of Product

Moving the solution beyond the POC to product with AI as core.

Launch Testing and Maintenance


Integrating the product developed with your existing platforms.

Decisions Insights AI and ML

Intelligence Process Automation for Insurance Claiming Process, that Understands/Analyses multiform Inputs such as Collision Images/Videos, Driving Licence, Insurance Certificates.

User Uploads the image of the accident to the Insurance Company on any of the messaging platform this is using SXP and the AI detects what type of accident, either it is self accident or third party involved in the accident and detects the Vehicle Number.

Then User authentication is done by the detecting the Driving Licence and the Insurance Certificate once done, AI creates an Insurance Claim ID and initiates the BPM process.

To improve the early detection of Keratoconus Progression (KP) on tomography images obtained from schiemplug imaging tomographer. This image-based deep learning system aims at detecting progression before the change in steep or flat Keratometry value of >1D (Keratometry Change-KC).

The tomography images from the patients’ eyes with their follow-up visits were included in the study. These images were classified as progressive (726) and non-progressive (702) based on a change in flat or steep Keratometry of 1D or < 1D in 2 consecutive visits respectively. Changes in sagittal curvature (SC), posterior elevation (PE), anterior elevation (AE), corneal thickness (CT) images from the scheimpflug device were used independently on a deep learning network developed in TensorFlow. Prediction accuracy, as well as early prediction, is computed for each model. Early predictions from three different perspectives were analyzed. The models performance in relation to, scenario-1: KC metric across all scans > 1D, scenario-2: among consecutive scans with KC metric <1D. scenario-3: stage in terms of best scan of the prediction in a series of scans were assessed. An ensemble of all the models is also considered.

Conclusions: Predicting early progression of Keratoconus is very important in the management of the disease to avoid irreversible changes in corneal curvature which in-turn affects the visual acuity. In our initial results using deep learning models from the scans obtained from scheimpflug device, progression was detected much earlier than a change in the Keratometry of > 1D which is commonly used as a clinical guide for progression.

We recently completed the Data/Terms Extraction Solution: The objective of this project was to design a solution which takes in a clean PDFs* and extracts aforementioned key terms and their corresponding values from the uploaded document. The purpose of this is to reduce manual labor costs by utilizing OCR and AI in the extraction of required terms and to reduce the time taken by the process.

Project success was defined as designing a solution/product which would contain extraction algorithms for set number of terms, provide the functionality to export all the values extracted for the terms into an excel, provide a minimalistic UI to work with. The project was started by defining Clean and Valid document as an input to the solution. A Clean document is one with No OCR issues such as no stamping’s, low or poor scan quality, no missing text. Also, It should not have scanning issues such as Orientation problem, multi column PDF. Also, Handwritten data, tables/images/shapes will not be extracted. A Valid document is one which contains keywords provided in Model. Also, It should be from the document types provided in the sample data. Cross reference documents are also not supported.

The deliverable for Lease Extraction Solution project phase 1 is the automatic extraction of 64 terms when provided with a clean and valid lease document with all features which were agreed as part of the scope.

It is no secret that the Fintech space is undergoing massive changes, especially in India. Traditional banking models have left a significant amount of people under served. Couple this with a complex web of regulations, and a slow uptake of technology, and this area is prime for disruption from ambitious Non-Banking Financial Companies (NBFCs).

Enter AbhiPaisa

AbhiPaisa is an innovative lending platform that aims to address the credit gap for the young salaried individuals. Based on extensive internal research, AbhiPaisa had concluded that most of these individuals are looking for bridge loans that get them through to the next salary. This could be for a payment of an EMI, or money for a flight ticket home or gifts during the holiday season.

These are individuals that have been left out of the current credit systems primarily due to a lack of credit history. The current models rely on credit history, transaction analysis and current worth to decide the eligibility for a loan. While these are no doubt significant parameters, they fail to paint an accurate picture given the small loan amounts and the short repayment periods. Typically, young individuals availing such loans are rejected by a lack of credit history, poor credit ratings and a lengthy loan approval/ disbursement process.

The Solution

So, the problem was clearly how do you get to service these young individuals without much of a credit history and provide them access to quick, easy bridge loans? Instead of trying to solve that problem, AbhiPaisa decided to instead redefine the problem to something far more ambitious. AbhiPaisa decided that they would offer bridge loans to salaried employees and the loan amount would be disbursed to the customer’s bank account in 5 minutes. As the technology partner, it was left to us to devise a risk-free model to enable this to happen.

Ability & Intent

Based on statements of transactions, it became a fairly simple to problem to arrive at the ability of the customer to pay. The more pressing problem was to discover the intent to pay. Without a history of repayments, how were we to predict whether the customer intends to pay?

The answer came through the recent developments in Artificial Intelligence and Machine Learning. Existing literature had successfully shown that social media information was an useful data pointer for predicting intent to pay.

Machine Learning

Put simply Artificial Intelligence (AI) is a computer system that mimics and/or replicates human intelligence. Machine Learning allows computers to learn on their own. Machine learning analyzes data and crunches numbers, learns from it, and uses that to make a prediction/truth/determination depending on the scenario. The machine is essentially being trained, or really training itself, on how to perform a task correctly after learning from all the data it has analyzed.

We started out with a base model with relevant parameters that were of interest. To obtain data to train our models, AbhiPaisa went ahead with the bold decision of giving out loans based on existing credit models. In fact, we even went ahead with issuing a few loans that against the recommendation of existing credit models. The results (defaults, successful payments) of all these loans were used to train the AI model. Once the model started predicting the results with reasonable accuracy, a second set of loans were disbursed. As expected, and much to the relief of the AbhiPaisa, the machine started outperforming the current credit models/ SMEs in the decision of issuing a loan.

It is fascinating that technology has evolved to a stage where it is able to allow us to get a look into the psyche of a person, something which SMEs are unable to do. We have been greatly encouraged by the results and are working on exciting problems that can be solved by AI & ML.

Start Conversation

start improving your business today

Contact our team if you have any further questions. We are here to help you out!

For a Demo