Electrical engineer. Machine learning researcher and engineer, based in Bolivia.

• Language as operator algebra. Words as unitary matrices, sentences as their ordered, non-commutative products — so word order is carried by the algebra itself, with no positional encodings.
• ML systems engineering at Deep Microsystems. Taking models from a research notebook to something that runs in production.
• Open-source reproducibility. The code behind the paper is public, with the raw logs the result tables are rebuilt from.

All You Need Is Non-Commutative Words Carla M. Quispe Flores*, Stanley Salvatierra*, Renan Cabrera — *equal contribution arXiv:2608.29314 [cs.CL], 2026 · abstract · PDF · code (MIT)

Tokens are represented as unitary matrices and a sentence as their ordered product. Because matrix multiplication does not commute, word order comes out of the algebra rather than a positional encoding. The same algebra also gives self-attention with no query, key, or value projections, a canonical-coset readout, and continual learning by nested group extension. Across standard text-classification benchmarks the method matches or exceeds bag-of-words baselines — replacing a ~30,000-dimensional vocabulary space with a dense, 64-parameter real-valued encoding.

This construction is inspired by the mathematics of quantum mechanics, but every computation here is classical and we claim no quantum advantage.

• Site: https://stanlee321.github.io
• X: @iamatachyon

PyTorch · Node · Go · gRPC · Kubernetes · Flutter/React · AWS

Solve the problem of intelligence, solve everything else!