When we started using Clickhouse, the concepts of MergeTree and SummingMergeTree was very confusing. Not only should one understand what each mean, it is extremely crucial to know when to use what as using a wrong one for the other would either result in Resolution/Data loss or would incur huge amount of storage space.

Materialized Views are not something very new if you are familiar with Database views and have worked with Time series/Analytical database. However, when used in Clickhouse, they are blazing fast. So fast, that not many databases in the similar domain can even come closer to it.

GCP is right up there in the top three race to cloud infrastructure only behind Amazon Web Services and Microsoft Azure. It has almost everything you need from classical VM instances, Serverless, Hosted Kubernetes, Docker container registries, ML, Vision - almost everything!

We had an interesting requirement to authenticate IoT devices by our cloud service for managing those devices, provisioning them, rotate keys, etc.

With the development community leaning a lot towards front-end, and framework like Rails mostly moving to mostly serve back-end, it is important to keep this separation clean and straightforward in a Production environment as well.

We have seen how to build a custom object detector using Faster RCNN with PyTorch. That model performed pretty decently with the IOU of 71%, this can further be improved with little tweaks in training and inference.

With so much advancement in technology and computing power, we are at ease in dealing with many of the supervised learning problems. This learning method requires tons of labeled data and we train our model on this data over and over to learn the pattern. This is effective and extremely performant. But we won’t be having the luxury of labeled data always. In many real-time cases, we will always have unlabeled data and need lots of time and effort to label this manually.

If you are familiar with object detection then it’s quite certain that you know a bit about YOLO. It’s an efficient and faster object detection algorithm and the first choice for real-time object detection tasks.

Object Detection is always a hot topic in computer vision and is applied in many areas such as security, surveillance, autonomous vehicle systems, and machine inspection. Widely used object detector algorithms are either region-based detection algorithms (Faster R-CNN, R-FCN, FPN) or single-shot detection algorithms(SSD and YOLO).

In one of the projects I worked on, I had written close to thousands (sic) of Regex as they help in having a strong foundation in Natural Language Processing besides the existing libraries available.

Refactoring is nothing but restructuring the existing code such that the existing behavior that is accessed through the outside world, remain the same. It’s often referred to as “under the hood” changes.

One of the sexy features of capistrano if you had used it with Rails is how super simple it is deploy a Rails project. The DSL (Domain Specific Language) it uses is very straightforward for even a beginner to understand. However, with time, Rails at least in our experience has moved mostly to serve the backend alone with the Front end being taken over by the likes of Vue, React etc.,