Module compression_semantic 

Semantic Structured Compression — RML-1208

Compresses verbose memory content into structured summaries targeting ~30x token reduction using rule-based NLP techniques. Pure computation — no database access, no network I/O.

Pipeline

1. Split text into sentences
2. Strip filler and hedging phrases
3. Extract proper nouns and number/date entities
4. Derive subject-verb-object cores
5. Deduplicate near-identical sentences (Jaccard > 0.6)
6. Reassemble structured_content and key_facts

Invariants

• Never panics on any input (including empty strings)
• Token estimation uses text.len() / 4
• Short content below min_content_length is returned verbatim

Structs

CompressedMemory

The result of compressing a single piece of text.

CompressionConfig

Configuration for the semantic compressor.

SemanticCompressor

Rule-based semantic compressor — no ML required.