GenCuit-AI is a miniature Small Language Model (SLM) built from scratch in PyTorch. Instead of predicting standard text, this customized Transformer architecture acts as a biological "autocomplete"—specializing in the design of functional, synthetic genetic circuits.
This AI was developed to serve as the backend computational logic for interactive microbiology learning modules (like MetaboQuest) and to model autonomous, multi-organism life-support pathways for Mars colonization architectures (LUMIDOMUS) under the TAIROE research initiative.
- Custom Transformer Architecture: Implements a full Self-Attention mechanism, token embeddings, and positional encoding from the ground up to understand the "grammar" of DNA.
- Multi-Domain Biological Vocabulary: Trained on a unified dataset that seamlessly bridges bacterial (E. coli sensors), plant (photosynthetic carbon-fixation), and human (mammalian expression) genetic sequences.
- Temperature Control: Features a dynamic creativity slider. Lower temperatures yield strictly conserved, logical operons, while higher temperatures simulate evolutionary mutation for discovering novel pathways.
- Interactive GUI: Wraps the complex PyTorch tensor math in a clean, user-friendly Streamlit web application.
- Visual Dictionary: Automatically parses the AI's mathematical output and breaks down the biological function of each generated gene (e.g., explaining
pCaborRuBisCO) for non-scientist users.
- Deep Learning Framework: PyTorch
- Frontend Interface: Streamlit
- Language: Python
- Data Processing: NumPy, Pandas
git clone [https://github.com/12415749/GeneCuit-AI.git](https://github.com/12415749/GeneCuit-AI.git)
cd GeneCuit-AI