LlamaTranslator

This is a system designed to translate Java Language Specification (JLS) sentences into Prolog code using the Llama3 as our language model and sentence-transformers/all-MiniLM-L6-v2 for generating embeddings.

The process flow is as follows:

1.) Receive the JLS sentence as our query

2.) Generate query embedding using the embedding model

3.) Perform Hybrid Search on the Knowledge Base

4.) Use RAG to retrieve relevant context and augment prompts

5.) Use APE for generating the prompt for expanding the query

6.) Feed to LLM to process the prompt to generate the expanded query

7.) The LlamaTranslator system generates a translation prompt using the expanded query and context from RAG

8.) The LLM processes the translation prompt to generate Prolog translation and identify intersecting predicates

9.) The knowledge base is updated and returns the information on translation with intersecting predicates

Planned improvements:

• Add evaluation metrics (TBD) for improving consistency and accuracy of JLS-to-Prolog translations

• Implement feedback mechanism for rating translations and for improving APE and RAG processes

• Incorporate cross-reference solution by aggregating information from referenced section in the JLS to the original statement