This release introduces an experimental but working transformer model in Rust, and adds a reference Axum-based model evaluation service. The genetic algorithm received a configurable speciation system and better documentation. Startup time improved significantly thanks to disk caching of n-gram models. A large number of AI-agent-assisted bug fixes, unit tests, and cleanups were applied across the codebase — greatly improving reliability. The frontend/backend stack was upgraded (Spring Boot 4.x + Angular deps), and WebSocket stability was fixed.

Transformer / GPT Model Implementation (core ML breakthrough)

• First working transformer model implemented
• Subtle bugs in the GPT/transformer model fixed
• Sampling logic cleaned up & moved to lib.rs, tokenizer whitespace removed

Rust Backend / Model Serving Infrastructure

• Added reference Axum service in Rust to evaluate transformer model outputs

Heavy Agentic / AI-Assisted Development Wave

• Many rounds of bug fixes & unit tests created or suggested by AI agents
• Documentation added/updated across all modules
• TODOs resolved throughout the codebase

Genetic Algorithm & Evolutionary Improvements

• Made speciation operator configurable + added RandomSpeciationOperator

Performance & Usability Enhancements

• N-gram model cache now written to disk → much faster startups
• Fixed WebSocket session closed errors

Dependency & Maintenance Updates

• Spring Boot upgraded to 4.x + npm/Angular dependencies refreshed

Photo by Igor Omilaev on Unsplash