Update README.md

README.md

@@ -12,64 +12,42 @@ emoji: 🌍
 
 # Universal Computing Research
 
-**Universal Computing Research** is an independent AI research organization focused on efficient, compact, and architecture-driven deep learning.
+Universal Computing Research is an independent research organization focused on compact language models, efficient architectures, tokenization, and training methods that improve capability under limited compute and parameter budgets.
 
-We build small language models, parameter-efficient neural layers, custom tokenizers, and research artifacts that test how far useful intelligence can be pushed under strict compute, memory, and parameter budgets.
+We release models, code, and experiments for open and reproducible research.
 
-## Research direction
+--- 
 
-Our work is centered on a simple question:
+## Model families
 
-> How much capability can be recovered through better architecture, tokenization, data curricula, and parameterization, without relying only on scale?
+### Limen
 
-Current focus areas:
+Limen is our family of small general-purpose language models.
 
-- Small language models
-- Parameter-efficient architectures
-- Random projection layers
-- Custom tokenization pipelines
-- Arithmetic and algorithmic reasoning
+The first release, **Limen0.2B**, is a 222.5M-parameter base model trained from scratch on 50B tokens. It uses a BoundlessBPE tokenizer that can learn tokens spanning whitespace, along with grouped-query attention and an experimental XSA projection.
 
-## Released models
+[View the Limen model](https://huggingface.co/UniversalComputingResearch/Limen0.2B)
 
-### Atom3.4m
+### Atom
 
-A 3.41M parameter decoder-only language model trained from scratch for studying compact architectures, curricula, and small-model benchmarking.
+Atom is our family of extremely small experimental language models.
 
-- Grouped-query attention
-- RoPE positional embeddings
-- RMSNorm
-- Gated SiLU feed-forward layers
-- Custom 4,096-token byte-level BPE tokenizer
-- Approximately 5B training tokens
+These models are used to study compact transformer architectures, custom tokenizers, arithmetic representations, and training curricula at parameter counts below four million.
 
-[View Atom3.4m](https://huggingface.co/UniversalComputingResearch/Atom3.4m)
+[View Atom3.4m](https://huggingface.co/UniversalComputingResearch/Atom3.4m) · [View Atom2.7m](https://huggingface.co/UniversalComputingResearch/Atom2.7m)
 
-### Atom2.7m
-
-A 2.74M parameter causal language model with an arithmetic-aware tokenizer and digit-structure features.
-
-- Custom byte-level BPE tokenizer
-- Atomic digit and operator handling
-- Least-significant-digit-first numeric representation
-- Place and role embeddings for integer arithmetic
-- Strong ArithMark-2.0 performance for its size
-
-[View Atom2.7m](https://huggingface.co/UniversalComputingResearch/Atom2.7m)
+---
 
 ## Research
 
 ### Parametrized Random Projection
 
-We study **Parametrized Random Projection** layers as lightweight replacements for dense linear layers.
-
-The core idea is to separate fixed feature mixing from learnable adaptation: a non-trainable random projection performs the mixing, while small learnable element-wise parameters modulate the input and output.
-
-This reduces trainable parameter count from quadratic to linear scale while preserving much of the utility of dense projections.
+Parametrized Random Projection replaces dense learned projections with fixed random feature mixing and small learned input and output parameters. The goal is to reduce trainable parameter growth from quadratic to linear while retaining useful projection capacity.
 
 [Read the paper](https://arxiv.org/abs/2512.13480)
 
-## Open source
+---
 
-Our models and research artifacts are released to support reproducible, open, and practical AI research.
+## Open source
 
+Our models, code, and research artifacts are released for reproducible and practical research.