We are pleased to announce that Nyckel Image Similarity Search API technology has been integrated into Eden AI API.
Nyckel is a company that provides rapid machine learning solutions for developers. Their efficient and secure API facilitates seamless integration of their machine learning services into your applications. It allows you to incorporate AI into your product without the need for a dedicated machine learning team or costly hardware investments.
With Nyckel, you can train your model in just five minutes, integrate it within ten minutes, and rely on their user-friendly UI and API to manage the challenging aspects. Nyckel simplifies all the intricacies of machine learning, allowing you to concentrate on your data and specific problems.
Eden AI offers Nyckel Image Similarity Search solutions on its platform amongst several other Computer Vision technologies. We want our users to have access to multiple AI engines and manage them in one place so they can reach high performance, optimize cost and cover all their needs.
There are many reasons for using multiple AI APIs :
You need to set up an AI API that is requested if and only if the main AI API does not perform well (or is down). You can use the confidence score returned or other methods to check provider accuracy.
After the testing phase, you will be able to build a mapping of AI vendors' performance that depends on the criteria that you chose. Each data that you need to process will be then sent to the best API.
This method allows you to choose the cheapest provider that performs well for your data. Let's imagine that you choose Google Cloud API for customer "A" because they all perform well and this is the cheapest. You will then choose Microsoft Azure for customer "B", a more expensive API but Google performances are not satisfying for customer "B". (this is a random example)
This approach is required if you look for extremely high accuracy. The combination leads to higher costs but allows your AI service to be safe and accurate because AI APIs will validate and invalidate each other for each piece of data.
We had the chance to talk to Oscar Beijbom, Nyckel’s Chief Technical Officer, who agreed to answer some of our questions:
I’m the co-founder and CTO at Nyckel. I have been doing various types of ML for 20 years, all the way from shipping products to research.
Having gone through ML product development several times in my career, I noticed a pattern in the toolchain. Regardless of the application, you need more or less the same stuff. For example, you need to be able to view, filter, and search your data. You need to annotate it. You then need a way to iterate on the ML method itself, which in turn requires a compute cluster, performance monitors and such. You then need to deploy the trained model. Etc. etc. None of these steps are super complicated, but there is a lot of room for mistakes and poor design choices.
At the same time, my co-founder Dan, who is an experienced developer, but not a ML person, had experienced the frustration of getting up and running with ML.
So we thought: why not package the whole thing up, end to end, and let people interact at the very highest level: tell the system what outputs they expect for each given input, and then (on the other end of the pipeline) provide a deployed elastic endpoint for their model.
We started Nyckel in 2021, and participated in the Y-Combinator in 2022
We are an API-first SaaS platform, but we also offer a comprehensive User Interface which is convenient for annotation and reviewing the data.
We pride ourselves in our “no toggles” approach to ML. Meaning: we don’t expect people to select the model, split the data, set hyperparameters, or internalize complicated machine learning metrics.
Customers also love our super-fast train-and-deploy time. Basically: within seconds of finishing annotation, your model is not just trained but also deployed to an elastic inference endpoint. Even with this fast turnaround time, our models are just as accurate as other top players.
Our most popular functions are custom image and text classification, but we also support semantic search, OCR, and other things.
We see a wide range of customers. Our ideal customers are high level technical folks from smaller companies. They can see the value right away and make a purchasing decision. Some common use-cases are content moderation, image classification workflows, and tagging, search for webshops.
Eden AI does a great job helping non-experts navigate the ML landscape. We are excited to be part of that ecosystem, and add more functionality to the platform over time.
In a few directions. First, we are constantly adding more input and output types including generative AI concepts like image generation from text.
Second, we started offering pre-trained (static) functions.
Finally, we are just rolling out a new integrated active learning feature which we call “invoke capture”. Invoke capture “captures” samples that are passing through the invoke endpoint and adds them to a staging area. The user can then label the samples directly in the UI. This allows them to continuously improve their function without having to write any code.
We think that AI powered decision making will continue to be the work-horse of process automation. Our goal is to be the AI engine that powers all of that.
You'll need some documentation to use Nyckel API on Eden AI.
First upload the images to create your dataset:
Then launch the image similarity search API:
Eden AI is the future of AI usage in companies. Our platform not only allows you to call multiple AI APIs but also gives you :
You can see Eden AI documentation here.