A complementary open-source chain.
Large language models produce inefficient tokens in most languages and fail on domain-specific questions. At Magibu we solve this at every link of the chain with open source: from tokenization faithful to a language's morphology, to a methodology for building embeddings in your own language; from high-quality fine-tune data, to tools that connect the model to the outside world.
Morphological Tokenizer
Split text into morphology-faithful units and recombine them.
Language-Native Embeddings
An open method for building tokenizers + embeddings in your own language.
Fine-tune Datasets
High-quality open datasets tailored to task and persona.
LLM Tools
The ability to call the right tool at the right time.
How do I contribute?
- 01Go to the repo you're interested in on GitHub and review open issues
- 02Comment on an issue or open a new one
- 03Fork the repo and create a new branch
- 04Make your change, test it, and document it
- 05Open a pull request - explain what and why
- 06After review, merge - join the contributor list
Open Source Projects
Turkey-focused open R&D. Pick up issues on GitHub or apply to join the team. These contributions are prioritized in future hiring.
Turkish Morphological Tokenizer
ActiveA modern tokenizer that splits text into morphological units faithful to Turkish phonetics and can recombine them.
Language-Native Embeddings
ActiveAn open methodology compiling methods and steps for anyone to build efficient tokenizers and embedding models for their own language and domain.
Fine-tune Datasets
ActiveA community project gathering high-quality Turkish/English dialogue datasets in a single format to adapt models to a task, language, or persona.
Magibu LLM Tools
ActiveTools and auxiliary systems the model uses at inference time - overcoming model limits by calling the right tool at the right time.
Community
Open science, benchmarks, and community contributions.
Open-Ended Live Stream
A Sunday live stream with no end time: code, questions, papers. Leave a request for the topics you want covered.
Request form →Magibu AI Weekly
Open-source weekly digest: AI news, papers, models, benchmarks, and underrepresented language updates.
View archive →