Plan for Clojure AI, ML, and high-performance Uncomplicate ecosystem in 2026

November 29, 2025

Please share this post in your communities. Without your help, it will stay burried under tons of corporate-pushed, AI and blog farm generated slop, and very few people will know that this exists.

These books fund my work! Please check them out.

I've applied for Clojurists Together yearly funding in 2026. Here's my application. If you are a Clojurists Together member, and would like to see continued development in this area, your vote can help me keep working on this :)

My goal with this funding in 2026 is to continuously develop Clojure AI, ML, and high-performance ecosystem of Uncomplicate libraries (Neanderhal and many more), on Nvidia GPUs, Apple Silicon, and traditional PC. In this year, I will also focus on writing tutorals on my blog and creating websites for the projects involved, which is something that I wanted for years, but didn't have time to do because I spent all time on programming.

How that work will benefit the Clojure community

This will highly benefit the Clojure community as this is THE AI ecosystem for Clojure, and supporting AI is arguably the main focus on probably all software platforms. Clojure has something to offer on that front, beyond just calling OpenAI API as a web service!

Uncomplicate grew to quite a few libraries (of which some are quite big; just Neanderthal is 28,000 lines of highly-condensed, aggresively macroized, and reusable code): Diamond ONNX Runtime, Neanderthal, Deep Diamond, ClojureCUDA, ClojureCPP, Apple Presets, ClojureCL, Fluokitten, Bayadera, Clojure Sound, and Commons.

Here's a word or two of how I hope to improve each of these libraries with Clojurists Together funding in 2026.

Neanderthal (Clojure's alternative to NumPy, on steroids)

In 2025, Neanderthal celebrated its 10th birthday. It started as a humble but fast matrix and vector library for Clojure, but after 10 years of relentless improvements, now it boasts a general matrix/vector/linear algebra API implemented by no less than 5(!) engines for CPUs, GPU (Nvidia CUDA), GPU (OpenCL: AMD, Intel, Nvidia), Apple Silicon (Accelerate), and general CPU (OpenBLAS). And this is not a superficial support for the sake of ticking a check box; each of these engines support much more operations on exotic structures, and configuration options, than I've seen elsewhere. It has almost everything, but it doesn't (YET!) have a METAL-based engine for Apple GPUs. Let's work on that!

Deep Diamond (the Clojure Tensor and Deep Learning library, not quite unlike PyTorch, but of different philosophy)

In 2019, I started Deep Diamond as a demo showcase for Neanderthal's capabilities as the foundation for high-performance number-crunching software. It quickly outgrew that, and became a general Tensor/Deep Learning API, implemented by several fast, native optimized, backends, that run on both CPUs and GPUs, across hardware platforms (x86_64, GPUs, arm64, Apple Silicon, you name it) and operating systems (Linux, MasOS, Windows). Of course, it does not clash with Neanderthal, but complements it, in the best manner of highly focused Clojure libraries that do one job and do it well.

Deep Diamond is quite capable, but it cries for a METAL-based engine for Apple GPUs, too.

Diamond ONNX Runtime (the Clojure library for executing AI models)

This is the latest gem in Uncomplicate's store, and I developed it thanks to Clojurists Together funding in Q3 2025. Similarly to how I started Deep Diamond mainly as a teaching showcase for Neanderthal, I started this to show Clojure programmers how close we, Clojurists, are to the latest and shiniest AI stuff that everyone's raving about. But of course, being close does not mean that we can close the gap to the multi-billion funded Python ecosystem in a few afternoons. It needs laser-focused development and knowing what to do, when, and where. Nevertheless, Clojure is there. Now we can run inference on the trained models from Hugging Face and other vibrant AI communities directly in Clojure's REPL. Does this make an effortless billion-dollar AI startup? NO. Does it bring Clojurians to the party? YES! And there's more to come.

Not only that this library is new, but the whole wider ecosystem exploded in the last year with the wide availability of open-weights model that you can run at home. So, lots of functionality is added upstream all the time, and I hope to be able to stay current and have the best and newest stuff in Clojure..

ClojureCUDA (REPL-based low-level CUDA development)

Not many Clojurians may prefer to work with GPU directly, or to write their own kernels. Neither do I. But, this library is one of the un-celebrated workhorses that enables me to implement whatever I want in Neanderthal, Deep Diamond, and Diamond ONNX Runtime, instead of just trying to wrap whatever there is in upstream C++ libraies. ClojureCUDA gives us the superpower of choice: wrap whatever works, but then implement the missing parts yourself!

As CUDA is receiving a steady stream of changes and improvements, I'd like to improve and extend ClojureCUDA to always be in top shape! It is not as easy as it seems to the casual onlooker.

ClojureCPP (the gateway to native C++ libraries)

From 20,000 feet, integrating a native library to JVM and Clojure may look straightforward. Oh, how wrong they are. Virtually every C++ library is a special kind of jungle, with its own structures, patterns and inventions. What might seems a minor technical detail might require special acrobatics to support it on the JVM. Masking that mess under the hood so that a Clojurian do not need to care might be insanely brittle if it weren't for ClojureCPP! It is not as large as Neanderthal or Deep Diamond, but it is one of the reasons that enables these upper level libraries stay on the 25,000 or 3,000 lines of code mark, instead of being 500,000 or 50,000, as many of their counterparts in other languages.

Apple Presets (native JNI bindings for various Apple libraries)

Yup. To support Apple Silicon in Neanderthal and Deep Diamond I had to make these bindings, since there weren't any to "just" wrap. And to support more Apple high performance computing apis, I'll have to create additional bindings (for example, for METAL) and only then develop the Clojure part.

Fluokitten, Bayadera, ClojureCL, Commons, Clojure Sound, etc.

These libraries will not be in focus in 2026., but will probably need some care and assorted improvements here and there.

Summary:

In short, my focus with this funding with Clojurists Together will have two main branches:

Development of new functionalities, supporting more hardware and platforms for existing functionality, and fixing issues for a dozen Uncomplicate libraries that already exist. This is what is described in the text you've just read.
Develop an unified website for Uncomplicate and stuff it with useful information in one place. Currently, some libraries have websites that I wrote many years ago, while some rely on GitHub Clojure tests, in-code documentation, tutorials on dragan.rocks and my books. There are many resources, some of which are quite detailed (2 full books!), but people without experience (which is the majority of Clojure programmers) have a hard time using all these in organized way. I hope to solve this with the unified website!