What problem does this solve?
Currently, the AAAK (Always-Available Asking Knowledge) extraction and entity detection are optimized for English. To make MemPalace accessible and effective for the Brazilian community, it needs to correctly identify patterns and entities in Portuguese (PT-BR) without losing accuracy in the memory hierarchy (Wings/Halls/Rooms).
What's the proposed solution?
I propose adding PT-BR localization support for AAAK. This includes:
Updating entity_detector.py or equivalent modules to recognize PT-BR patterns.
Ensuring the "Verbatim First" principle is maintained by storing exact Portuguese terms.
Keeping the implementation "Zero API by default" and "Local first", running entirely on the user's machine.
Adding specific test cases in tests/ for Portuguese inputs to ensure no regressions in retrieval (aiming to maintain the 34% improvement driven by the Palace structure).
Alternatives considered:
Using external translation APIs (rejected as it violates the "Zero API" and "Local first" principles).
Manual tagging (less efficient than improving the automated detector).
What problem does this solve?
Currently, the AAAK (Always-Available Asking Knowledge) extraction and entity detection are optimized for English. To make MemPalace accessible and effective for the Brazilian community, it needs to correctly identify patterns and entities in Portuguese (PT-BR) without losing accuracy in the memory hierarchy (Wings/Halls/Rooms).
What's the proposed solution?
I propose adding PT-BR localization support for AAAK. This includes:
Updating entity_detector.py or equivalent modules to recognize PT-BR patterns.
Ensuring the "Verbatim First" principle is maintained by storing exact Portuguese terms.
Keeping the implementation "Zero API by default" and "Local first", running entirely on the user's machine.
Adding specific test cases in tests/ for Portuguese inputs to ensure no regressions in retrieval (aiming to maintain the 34% improvement driven by the Palace structure).
Alternatives considered:
Using external translation APIs (rejected as it violates the "Zero API" and "Local first" principles).
Manual tagging (less efficient than improving the automated detector).