This is a introduction to Oculow, a startup with hopes of achieving great advances in the use of A.I and software Q.A. We are currently 3 enthusiastic developers looking to generate a impact on the software development community by using A.I to create leading edge software QA tools. Any feedback you have for us is greatly appreciated, no matter if it’s positive or negative!
What is oculow
Oculow is a platform that allows users to detect visual errors in their application/website through the use of A.I. Think of this as a spell check but for visual errors in your app instead of words. This is something that will help developers validate that their app is being rendered in all screen resolutions and sizes, allowing them to release new versions with confidence.
Besides the innovative error detection, we also provide the traditional difference between versions comparison. This is where users can define a baseline to then compare the new version of the system to the previous one, detecting any differences and failing the test if some are found. Then having the possibility of defining the new version as the new baseline.
By combining both features, the comparison logic and the error detections, we can handle the definition of baselines. This means that we allow developers to effortlessly scale their visual regression tests to many different resolutions while keeping the benefits of the tools that exists today.
In the error detection algorithm, we are currently detecting patterns that may represent overlapped elements. We want to get a good prediction percentage in this detection before moving on to detecting a different visual error. The biggest problem we face, along with any other machine learning projects, is the detection of false positives or false negatives.
When we talk about a prediction being a false positive, we mean any errors that are detected which are not actually errors. These are not as important, since the user will have the final say if there is any error or not.
The most important part of the detection we want to improve, is lowering the false negatives. By that, we mean any errors that are not detected and that we are training the model to check for. In order to improve this, we need more data to train upon. This will allow us to generalize better and detect on different types of sites. Not just from the data set we currently have.
To solve both of the issues above, we improve the accuracy of the model by feeding our neural network more data. This allows the algorithm to generalize better and gives the benefit of better detection to those who use the system more frequently. We can also look into redesigning our neural network if we see that with a large amount of data, no progress is made.
The core of our business will be the detection of defects in apps to assist users. For this reason, we want to prioritize our algorithm and improve its prediction accuracy before any other tasks.
We already have our product, but we need more people to find it and use it. We will be working on a launch campaign to get the attention of more developers and get some feedback from early adopters. Based on the feedback, we will re-prioritize our tasks if its needed.
If we start to get some more traction, we will be expanding our team with hopes of speeding up the process to a fully functional autonomous testing tool. Currently we are 3 developers handling everything, from marketing to the development of the core algorithm. If we can hire someone to handle all communications and we can focus on our primary goal, we can greatly increase development speed.
What we strive towards
We want to be the leading company that applies A.I. to resolve problems that developers may encounter when running Q.A. on their system. For that we have a long road ahead, but we are very certain that it is achievable. We believe were at the right start and we can only improve from here.
Once we are satisfied with our error detection algorithm, we want get into automated accessibility testing. We feel that this is an area that isn’t really looked into very often, and there are many people which suffer disabilities which would benefit from this. We can reuse a lot of our already trained neural network to generate some interesting feedback for developers.