Nyckel

What do you get when you combine Nyckel, Contextual.io, and OpenAI? An easy way to upload your dog's photo, discover its most-similar breed, and get additional fun facts about that breed. While this particular example is playful, similar workflows can solve real business problems. If you're using Contextual & Claude to analyze images, for instance, you could first run your images through Nyckel's Blurry Image pretrained classifier, then send Claude all the crisp images, saving on tokens and (perhaps) odd responses if Claude can't fully discern what's happening.

"There's an insanely broad range of custom classification use cases. Our customers constantly surprise us by what they build." - Oscar Beijbom, Nyckel's CTO on Episode 20 of 🧪Matthew Mirman, PhD's Accelerometer podcast. The podcast covers a variety of topics, including: - How many training samples are needed for an effective classifier - How generative AI is tackling classification - How businesses are leveraging custom classification to become more competitive You can watch the whole podcast here - https://lnkd.in/e6aNfA_M

🚀 Looking to build an intent classification engine? LinkedClient did just that using Nyckel’s classification API to build a custom model for analyzing sales inquiry responses. They tagged responses like "Interested" or "Neutral" with as few as 10 samples per class. In just minutes, they built an engine that has now classified over a million messages. You can read more about their journey here: https://lnkd.in/eGmWG9rP #AI #MachineLearning #SalesAutomation #Innovation #Nyckel #LinkedClient

How many training samples do you need for classification? Not as many as you might think. It's a question we often get, so our co-founder Oscar Beijbom analyzed 10,000 classifiers to find the answer. The number surprised even us, with the mean accuracy crossing ~80% after just 13 samples. This held true across industries and for both text and image classification. The article below dives deeper into the methodology and results. https://lnkd.in/eirww5Kc

Nyckel has a new model type! It’s called multimodal classification, and it allows for models trained on both image and text data (previously it was one or the other).  A retailer, for instance, could use a product’s image and name to autotag it to the right subcategory. This new function type adds to Nyckel’s suite of models, including multiclass and multilabel classification, image and text search, OCR, box detection, and center detection. If you’re interested in testing it, send us a note, and we'll enable it. https://lnkd.in/g2iFV9tf

Proud to announce that Nyckel has officially partnered with Eden AI to power their custom image classification offering. Eden AI customers can now build bespoke image classification models via an Eden AI API (which are then powered by Nyckel in the backend). If you haven't heard of Eden AI, check them out - they've built a great repository of AI/ML APIs for most any task. https://lnkd.in/dm_6D_CF

Image classification use cases almost feel limitless when you understand classification’s full scope & potential. For example, let’s say you’re in: → Retail: Enhance product categorization & customer search experience by tagging photos with automation. → Brand management: Detect the use of your logo across digital content. → Healthcare: Aid in medical imaging & diagnostics. Or what about the automotive industry? In a recent article, we share an illustrative example of how a car dealership could use image classification to label their vehicle inventory by brand, make, model, and/or color — creating big operational efficiencies. See just how quick & easy it is for a dealership to create an accurate image classifier with Nyckel. ⬇️ https://lnkd.in/gcmU22Qi #automotiveindustry #dealership #imageclassification #imagerecognition

It’s past time to say goodbye to irrelevant DAM tags, often created through generic auto-tagging APIs. These generic auto-taggers 1) don’t always give you the tags that you care about and 2) add noise by including tags that aren’t relevant. Instead, custom auto-tagging (which can be implemented in hours, not weeks!) can be trained on just a handful of manual tags — giving you the exact tags you need for your use case. So, what’s holding people who work in digital asset management back from using custom auto-tagging? Unfortunately, there are 2 big misconceptions: 1️⃣ You need a lot of data to train a custom model. In reality, modern models can be trained with just a tiny amount of data. In fact, we saw ~75% accuracy with just five examples per tag. 2️⃣ You need a team of AI experts to implement custom tagging. Nope. Simple and intuitive platforms like Nyckel now exist, ready for anyone to use. Link in the comments to see how our co-founder, George Mathew, trained and deployed a custom model to detect telehandlers in under three minutes. #DAM #digitalassetmanagement #imageclassification #imagerecognition

How do you make sense of massive amounts of remote sensing data when manual review isn’t feasible or practical? 🤔 ➡️ Enter: Image classification In our most recent article, learn how to use image classification to analyze your remote sensing data and how to quickly & easily build an image classifier with Nyckel. As an example, we show you how to solve Kaggle's Iceberg vs. Ship challenge using Nyckel. Find the link in the comments & let us know what you think. ⬇️ #remotesensing #imageclassification #environmentalmonitoring #agriculture #forestry #urbanplanning #disastermanagement 

Have you tried to use PyTorch or Keras to classify images, but struggled with the ML complexities involved? 😕 Good news: There’s an easier way. In two recent articles, we break down how to build image classifiers in both PyTorch and Keras — and then present an easier & quicker alternative (Nyckel 😏). Find the links to both articles in the comments. #pytorch #keras #imageclassification #imagerecognition