Voyance Vision
Aanuoluwapo Sebiomo
Voyance Vision is a computer vision solution that provides an automated, fast, accurate platform that is able to handle processing limitations such as low image
Voyance Vision is a Document Management System (DMS) that employs computer vision, Machine Learning and Artificial intelligence using trained models to extract, store, manage and retrieve documentations that businesses use daily.
Data Management Systems has become so crucial that its global demand reached $5.55billion in 2022, projected to reach $16.42 billion in 2029.
Vision can handle processing limitations such as low image quality or poorly scanned documents. It also automates the process so businesses can spend less time organising and retrieving files and more time working on other essential tasks.
My Role
I led the design of the project between February 2021 and November 2021. I designed all significant deliverables - prototypes, research documents, copies, design strategy, vision and design principles - this aided the different teams in aligning, especially during decision-making.
I worked alongside our diverse in-house teams - project managers, data scientists and engineers, frontend and backend developers and machine learning engineers. I also worked directly with the CEO who is a Software and Machine Learning Engineer.
Research and Insights
I worked with our project managers to research the problems that businesses face while handling documents and also the industries facing these challenges.
Voyance Vision would eventually have a wide range of use cases. For the sake of time and engineering limitations, the project managers and I decided to focus on one primary industry - the fintech space.
Our target was to create a solution that handles documents for KYC processes.
The challenge within the Fintech space
The FinTech space is a fast-growing niche in Nigeria, with each company having thousands of users. According to TechCabal, Kuda bank has about 650,000 customers, Paystack has about 60,000 customers according to TechCrunch, and CowryWise has over 220,000 customers (TechCrunch); these are only a fraction of the companies in Nigeria.
As a part of the compliance process, which ensures the authenticity of each customer and the overall integrity of the platform, customers need to submit any government-issued ID such as Driver’s license, Nation ID Card (NIN), Passport, Voter’s Card and or CAC Documents for businesses.
It can be a daunting task to go through the tons of documents submitted each day manually. These manual processes of extracting the needed data from these documents can be expensive, time-consuming, and error-prone.
The high-level goal
Our goal was to create a solution allowing businesses to process these documents effectively. Our research revealed that we could divide our high-level goals into five different sub-goals:
Speed
Accuracy
Automation/Autonomy
Cost
Security.
To compete with the developing DBS market, we needed to figure out how our product would meet the needs of the target users.
Improving speed and accuracy via continuous learning AI
The model's learning capability ensures that the machine can keep improving over time, making it better and more efficient. This learning ability enables it to understand and accurately capture data each time new documents are processed.
Handling processing limitations
The platform can accurately process documents despite limitations such as low image quality or poor lighting conditions.
Flagging tampered documents
Vision can also flag documents with low integrity or those tampered with due to manipulations.
What would success look like
Before beginning any design process, I wanted to define the metrics we would use to measure success.
From our high-level goals, I came up with four metrics -
The processing time
The processing accuracy
The rate of human intervention
The cost of the overall process.
I did not list security as a metric as it would be ingrained from the bottom up.
Refining the Problem Statement
African businesses need to quickly and accurately verify user documents with little or no human intervention and at an affordable rate.
Setting the design direction.
The project involved a cross-functional team - design, product, engineers, data scientists, and machine learning engineers. I was responsible for bringing all the individual units together to achieve the goal. It was a challenge having to collaborate with a diverse team as I needed ideas and buy-ins from the different teams all working remotely.
I approached this by dividing the project into two phases - the first phase would cover speed and accuracy, while the second phase would cover cost and automation. This allowed the product managers to effectively break down the tasks into smaller units with more focused goals.
The project's complexity meant that I needed to figure out almost everything before the teams would commit to building. The different teams needed to see a tangible document - this would save the headache of a back and forth.
Rapid prototyping was the fastest and most effective way to gain meaningful insights, feedback from the different teams, and management approval. These prototypes also helped to test and validate our solution with our target audience.
Voyance Vision
Vision is a cloud-based tool by Voyance that helps African businesses solve computer vision problems related to document and image analysis.
Train your models or use pre-trained models
Voyance Vision allows companies to train their models or use our pre-built models that our Data Engineers have trained. Pre-trained or prebuilt models help you extract essential information from documents without gathering data and then build and train your models.
Using pre-trained models is an excellent option for companies that need to conduct an analysis quickly and don't have the time or resources to build their models.
Seamlessly upload documents and export data.
Documents can be uploaded easily from local and cloud storage without disrupting your workflow. Data extracted can then be downloaded in various formats.
Annotate and Train Models
Vision allows its users to annotate documents - this is a process of labelling the data accessible in various layouts like video, text, or images. This process also improves accuracy.
Workflow Integration
Workflow integration reduces the rate of human interaction, instructing the system on what to do to extract data during certain events.
Testing our initial prototype
We got a few Data Scientists, Data Engineers and Machine Learning from outside our company to test our prototype. We observed that we did not give an option to upload PDF documents; this would come in handy for users and businesses with their documents saved as PDFs, not just images. We had assumed that the documents submitted would be in image format, but we were wrong.
A feature also requested was to be able to use the trained models as an API.
This feedback gave us much-needed insight and allowed us to iterate and include these features quickly.
The launch
On the 11th of August 2021, the project launched. The entire company was joyous as everything went seamlessly well. The hard work paid off.
The impact
Vision reduced manual intervention requirements by 80%, resulting in a 70% reduction in human resources dependence and a 90% reduction in processing time. Voyance Vision processed documents quickly and accurately without errors that may cause delays or loss of revenue. It also helped customers feel delighted with instant approvals and activation.
The Technologies We Used.
Our Engineers used:
Tensorflow, an open-source library for machine learning, mostly used for deep learning, was created by the Google Brain team.
Pytorch, an open-source library for machine learning, is commonly used for Computer Vision and Natural Language Processing.
OpenCV and
Kubernetes, a software orchestration tool.
Future use cases for Voyance Vision
Fraud detection: Vision can identify fraudulent documents, fake images, and other forms of invalid data. Vision can also detect unstructured data such as messages, claims, and customer feedback and then notify humans about incidents of suspected fraud. In addition to these benefits, Vision improves forecast accuracy, allowing loss control units to cover more cases with fewer false positives.
Vehicle damage assessment: Assessing damage to a vehicle takes much time and human resources. Vision can scan and analyse the damage to a vehicle in seconds. This allows the assessor to determine which vehicle elements should be fixed or replaced based on information from manufacturers.
Claims management: Machine learning and artificial intelligence are revolutionising the insurance industry, allowing businesses to automate claim management, detect incremental fraud patterns, and minimise false positives that differentiate between real and fake data points.
Challenges we Faced
The major challenge was identifying the specific solution we would offer at launch. Because computer vision is so robust, you can do a thousand things with infrastructure like Vision. So my work was to cut out many parts to create what our “right-now” clients need. Also, what we are building, an AI infrastructure, is more complex than your everyday tech product. I had to learn fast and have a clear picture of the product to help the team understand it. I am still doing the learning aspect because people are always coming to me with questions, and I should have the answers:- Jennifer Okafor, Product Manager.
The disparity in understanding of the problem was also a challenge. This was tackled by sharing concise product documentation across the team and having regular sessions with the product team to avoid assumptions. Formalisation of the engineering tasks, going from product description to an engineering breakdown. Architectural decisions had to be made, options had to be weighed carefully. - Temi Babalola, Senior Software Engineer.
Working on this project, I did not have background knowledge in OCR technology, Computer vision, Machine learning, and Artificial Intelligence. I had to study and research to develop an intuitive design that solved the problem. Working with the team on this project was terrific. - Sebiomo Aanuoluwapo, Product Designer.
The information in this case study is my own and does not reflect the views of Voyance. I have intentionally omitted confidential data where necessary.
Like this project
Posted Nov 16, 2022
Voyance Vision is a computer vision solution that provides an automated, fast, accurate platform that is able to handle processing limitations such as low image
Likes
0
Views
101
