Small steps.
A considered direction.
Build a useful foundation, test it with people, and grow from evidence. This roadmap separates implemented behavior from future work.
Working today · Development implementation
A native foundation.
- C++ interpreter with static checking and inferred return types
- Generic classes, private fields, and managed object memory
- Built in List, Dictionary, and Set with stable numeric List sorting
- Typed fixed and dynamic arrays with native vector loops and sorted search
- Typed matrices and ranks 1 through 8 with reshape, broadcasting, and matrix products
- Eight modules for data science and development
- Dense neural networks on CPU or available native MPS/CUDA; KNN and K means on CPU
- Explicit Float64/Float32 Tensor precision, GPU transfers, and synchronized benchmark checks
- Offline Iris and Wine with attribution and reproducible partitions
- Typed, budgeted CSV processing and persistent VS Code notebooks
- Development reports, notebook variable inspection, and fitted model checkpoints
- Optional native Aner kernel for real Jupyter notebooks
02
Next milestone · Public preview preparation
Ready to share and learn.
- Finalize the public repository and open source license
- Validate release packages and compatibility on each target OS
- Publish checksums, release notes, and contributor guidance
- Improve diagnostics and test real research workflows
- Develop tutorials and collect feedback from early users
03
Looking ahead · Design & research
Grow with the questions.
- Native compilation of Aner programs and execution backends
- Additional precision formats, GPU operations, multiple device scheduling, and distributed execution
- Strided array views, axis reductions, advanced indexing, and batched matrix products
- Richer table operations and additional data sources
- Streaming training observations and source linked pause/step debugging
- CNNs, RNNs, additional optimizers, SVMs, and decision trees
- Independent packages and a library ecosystem for outside contributors