VigenereSolver-ng

https://github.com/supasuge/VigenereSolver-ng

Description

VigenerSolver-ngis an experimental Vigenère solver that fuses classical statistics, beam-search heuristics, and modern language-model scoring. It can recover keys, decrypt ciphertexts, visualise key-length evidence, and benchmark different decoder backends — including KenLM language model. The previously known best language model prior to the release of the transformer architecture of neural networks powering modern LLMs today.

Project Goals

• [x] Fully decrypt a vigenere ciphertext with 0 knowledge of the key.
• [x] Understand and implement classical attacks defined in various research papers over time against the vigenere cipher.
• [x] Utilize modern-ish Language Model/NLP processing-based techniques to improve the overall accuracy.

TODO’s and Contributing

• [ ] Add particle optimization swarm markov chains
• [ ] Improve data normalization/pre-processing and post-processing.
• [ ] Switch from a LN-LM Model to a more modern architecture trained on a large corpus of english text first, then fine-tuned for the vigenere cipher.