@startuml Document Processing System Architecture
!include shared_interfaces.puml
package "Document Processing Core" {
abstract class BaseDocumentPlugin {
# chunkSizeTarget: int
# chunkOverlap: int
# sqlite_store: SQLiteStore
+ {abstract} createDocumentChunks(content): Chunk[]
+ {abstract} extractDocumentStructure(content): Structure
+ indexFile(path, content): IndexShard
+ generateDocumentSymbols(structure): Symbol[]
# optimizeChunkSize(content): int
# preserveSemanticBoundaries(text): string[]
}
interface IDocumentProcessor {
+ processDocument(content): ProcessedDocument
+ extractMetadata(content): Metadata
+ createSearchableChunks(content): Chunk[]
}
interface IChunkStrategy {
+ chunk(content, structure): Chunk[]
+ validateChunk(chunk): bool
+ mergeSmallChunks(chunks): Chunk[]
}
interface IStructureExtractor {
+ extractStructure(content): DocumentStructure
+ findSections(content): Section[]
+ buildHierarchy(sections): Tree
}
class DocumentChunk {
+ id: string
+ content: string
+ type: ChunkType
+ metadata: ChunkMetadata
+ embedding: float[]
+ contextBefore: string
+ contextAfter: string
+ bm25Terms: Map<string, float>
+ contextualEmbedding: float[]
}
class ChunkMetadata {
+ documentPath: string
+ sectionHierarchy: string[]
+ chunkIndex: int
+ totalChunks: int
+ hasCode: bool
+ language: string
+ keywords: string[]
}
class DocumentStructure {
+ title: string
+ sections: Section[]
+ metadata: Metadata
+ outline: TreeNode
+ crossReferences: Reference[]
}
enum ChunkType {
HEADING
PARAGRAPH
CODE_BLOCK
LIST
TABLE
METADATA
}
}
package "Document Plugins" {
class MarkdownPlugin {
- markdownParser: MarkdownParser
- sectionExtractor: SectionExtractor
+ createDocumentChunks(content): Chunk[]
+ extractDocumentStructure(content): Structure
+ parseMarkdownElements(content): Element[]
}
class PlainTextPlugin {
- nlpProcessor: NLPProcessor
- paragraphDetector: ParagraphDetector
+ createDocumentChunks(content): Chunk[]
+ extractDocumentStructure(content): Structure
+ inferDocumentStructure(content): Structure
}
}
package "Shared Components" {
class ChunkOptimizer {
+ optimizeChunks(chunks): Chunk[]
+ balanceChunkSizes(chunks): Chunk[]
+ ensureContextOverlap(chunks): Chunk[]
+ validateSemanticCoherence(chunk): bool
}
class MetadataExtractor {
+ extractFrontmatter(content): Metadata
+ inferMetadata(content): Metadata
+ parseYAML(content): dict
+ parseTOML(content): dict
}
class SemanticDocumentChunker {
- embedder: Embedder
+ chunkBySimilarity(content): Chunk[]
+ measureSemanticDistance(c1, c2): float
+ findOptimalBoundaries(content): int[]
}
}
package "Integration Layer" {
class DocumentAwareDispatcher {
+ routeDocument(path): IPlugin
+ handleDocumentQuery(query): SearchResult[]
+ extractSectionQuery(query): SectionQuery
+ selectSearchStrategy(query): SearchStrategy
}
class DocumentSearchEnhancer {
+ enhanceQuery(query): EnhancedQuery
+ parseNaturalLanguage(query): QueryIntent
+ findRelevantSections(query): Section[]
+ rankDocumentResults(results): Result[]
+ applyHybridSearch(query): HybridResult[]
}
class AdaptiveIndexer {
- chunkStrategies: Map<string, ChunkStrategy>
+ indexDocument(doc: Document): IndexResult
+ selectChunkStrategy(doc): ChunkStrategy
+ generateContextualEmbeddings(chunks): float[][]
+ updateBM25Index(chunks): void
}
}
' Relationships
MarkdownPlugin --|> BaseDocumentPlugin
PlainTextPlugin --|> BaseDocumentPlugin
BaseDocumentPlugin ..|> IDocumentProcessor
BaseDocumentPlugin ..|> IPlugin
BaseDocumentPlugin --> DocumentChunk : creates
BaseDocumentPlugin --> DocumentStructure : extracts
DocumentChunk --> ChunkMetadata : contains
MarkdownPlugin ..> IChunkStrategy : implements
PlainTextPlugin ..> IChunkStrategy : implements
ChunkOptimizer --> DocumentChunk : optimizes
MetadataExtractor --> BaseDocumentPlugin : used by
SemanticDocumentChunker --> BaseDocumentPlugin : enhances
DocumentAwareDispatcher --> BaseDocumentPlugin : routes to
DocumentSearchEnhancer --> DocumentAwareDispatcher : enhances
AdaptiveIndexer --> BaseDocumentPlugin : processes
AdaptiveIndexer --> ChunkOptimizer : uses
AdaptiveIndexer --> SemanticDocumentChunker : uses
' External Integration
BaseDocumentPlugin ..> SQLiteStore : persists to
BaseDocumentPlugin ..> SemanticIndexer : indexes with
DocumentAwareDispatcher ..> EnhancedDispatcher : extends
AdaptiveIndexer ..> ContextualEmbeddingEngine : generates embeddings
AdaptiveIndexer ..> BM25Index : updates
@enduml
