Mostlylucid.LucidRAG.DoomSummarizer.Core
2.1.0
See the version list below for details.
dotnet add package Mostlylucid.LucidRAG.DoomSummarizer.Core --version 2.1.0
NuGet\Install-Package Mostlylucid.LucidRAG.DoomSummarizer.Core -Version 2.1.0
<PackageReference Include="Mostlylucid.LucidRAG.DoomSummarizer.Core" Version="2.1.0" />
<PackageVersion Include="Mostlylucid.LucidRAG.DoomSummarizer.Core" Version="2.1.0" />
<PackageReference Include="Mostlylucid.LucidRAG.DoomSummarizer.Core" />
paket add Mostlylucid.LucidRAG.DoomSummarizer.Core --version 2.1.0
#r "nuget: Mostlylucid.LucidRAG.DoomSummarizer.Core, 2.1.0"
#:package Mostlylucid.LucidRAG.DoomSummarizer.Core@2.1.0
#addin nuget:?package=Mostlylucid.LucidRAG.DoomSummarizer.Core&version=2.1.0
#tool nuget:?package=Mostlylucid.LucidRAG.DoomSummarizer.Core&version=2.1.0
Mostlylucid.DoomSummarizer.Core
Core signal extraction, retrieval, and synthesis pipeline for news and content analysis.
Features
- Signal Extraction: Wave-based analysis pipeline (entities, topics, knowledge graphs)
- LLM Routing: Budget-enforced, circuit-breaking multi-provider routing (Ollama primary, cloud providers disabled by default)
- Relevance Scoring: 6-signal RRF fusion with query-type adaptive ranking
- Content Extraction: SmartReader + Markdown analysis for web content
- Knowledge Graphs: Entity extraction and relationship mapping
- Resilience: Circuit breakers, API budget tracking, adaptive rate limiting
- Local-First: ONNX embeddings, DuckDB vector search, Lucene.NET full-text search
Key Services
Retrieval Pipeline (RetrievalPipeline.cs)
Three-layer retrieval with parallel execution:
- Lucene.NET FTS - BM25F with field weighting (title 2x, keywords 2.5x, content 1x), Porter stemming, fuzzy matching
- Embedding HNSW - 384-dim cosine similarity with max-sim for composite queries
- Entity Profile HNSW - TF-IDF-confidence weighted entity fingerprints
Lucene and HNSW layers execute concurrently via Task.WhenAll, with results fused through RRF.
Synthesis (OllamaService.SynthesizeSummaryAsync)
Unified synthesis engine used by both scroll and ask:
- Smart evidence budgeting - per-item character budgets proportional to relevance; short items donate surplus to long ones
- TextRank compression - PageRank-style sentence centrality extraction using batch ONNX embeddings (no LLM needed)
- Semantic re-ranking - batch cosine similarity against the query for evidence ordering
- Clean prompts - evidence headers contain only sequential numbering and title; no metadata leaks to the LLM
Batch Operations (ItemProcessor.cs)
Optimized for throughput:
- Batch embedding - single ONNX forward pass for N items via
EmbedBatchAsync - Batch anchor computation - sentiment + topic anchors computed in one call at startup
- Batch indexing -
IndexBatchAsyncwraps save + FTS5 index + keyword corpus update in a single SQLite transaction
TextRank Extraction (TextRankExtractor.cs)
Deterministic key-sentence extraction without LLM:
- Sentence tokenization and embedding (batch ONNX)
- Cosine similarity graph construction
- PageRank iteration for sentence centrality
- Top-K sentence selection within character budget
Storage (StorageService.cs)
SQLite-backed storage with:
- Items, embeddings, FTS5 index, query log, URL cache
- Entity tables (entities, mentions, relationships)
- Batch operations for high-throughput ingestion
- Thread-safe semaphore-protected database access
Dependencies
- Mostlylucid.DocSummarizer - Shared NER, content extraction, RRF scoring, wave/signal types
- Mostlylucid.Summarizer.Core - Foundation pipeline interfaces
License
MIT
| Product | Versions Compatible and additional computed target framework versions. |
|---|---|
| .NET | net10.0 is compatible. net10.0-android was computed. net10.0-browser was computed. net10.0-ios was computed. net10.0-maccatalyst was computed. net10.0-macos was computed. net10.0-tvos was computed. net10.0-windows was computed. |
-
net10.0
- AngleSharp (>= 1.4.1-beta.506)
- DuckDB.NET.Data.Full (>= 1.4.4)
- Fluid.Core (>= 2.31.0)
- Markdig (>= 0.44.0)
- Microsoft.Extensions.Logging.Abstractions (>= 10.0.2)
- Microsoft.ML.OnnxRuntime (>= 1.23.2)
- Microsoft.ML.Tokenizers (>= 2.0.0)
- Microsoft.Playwright (>= 1.58.0)
- Microsoft.Recognizers.Text.DateTime (>= 1.8.13)
- Microsoft.Recognizers.Text.Number (>= 1.8.13)
- Microsoft.Recognizers.Text.Sequence (>= 1.8.13)
- Mostlylucid.LucidRAG.Decomposer (>= 2.1.0)
- Mostlylucid.LucidRAG.DocSummarizer (>= 2.1.0)
- Mostlylucid.LucidRAG.DocSummarizer.Anthropic (>= 2.1.0)
- Mostlylucid.LucidRAG.DocSummarizer.FullText.Lucene (>= 2.1.0)
- Mostlylucid.LucidRAG.DocSummarizer.FullText.Sqlite (>= 2.1.0)
- Mostlylucid.LucidRAG.DocSummarizer.OpenAI (>= 2.1.0)
- Mostlylucid.LucidRAG.DocSummarizer.Rdbms.Sqlite (>= 2.1.0)
- Mostlylucid.LucidRAG.LLM (>= 2.1.0)
- Polly.Core (>= 8.6.5)
- SmartReader (>= 0.11.0)
- Spectre.Console.Cli (>= 0.53.1)
- YamlDotNet (>= 16.3.0)
NuGet packages (21)
Showing the top 5 NuGet packages that depend on Mostlylucid.LucidRAG.DoomSummarizer.Core:
| Package | Downloads |
|---|---|
|
Mostlylucid.LucidRAG.Plugins.Books
Book summarizer plugin for DoomSummarizer. Provides hierarchical splitting (chapters, acts, sections), book type detection, and chapter-aware summarization strategies for novels, plays, anthologies, and academic works. |
|
|
Mostlylucid.LucidRAG.Plugins.Image
Image analysis plugin for DoomSummarizer. OCR, captioning, and multi-wave ML analysis. |
|
|
Mostlylucid.LucidRAG.Sources.Google
Google News and Places source plugin for DoomSummarizer. |
|
|
Mostlylucid.LucidRAG.Plugins.Video
Video analysis plugin for DoomSummarizer. Shot detection, scene segmentation, and transcription support. |
|
|
Mostlylucid.LucidRAG.Plugins.Data
Data analysis plugin for DoomSummarizer. Statistical profiling, schema detection, and constraint validation for tabular data. |
GitHub repositories
This package is not used by any popular GitHub repositories.
| Version | Downloads | Last Updated |
|---|---|---|
| 2.5.0-alpha0 | 402 | 2/10/2026 |
| 2.1.0 | 351 | 2/9/2026 |
| 2.1.0-preview2 | 345 | 2/9/2026 |
| 2.0.1-rc0 | 378 | 2/9/2026 |
| 1.1.1 | 420 | 2/4/2026 |
| 1.0.0 | 413 | 2/4/2026 |
| 1.0.0-rc6 | 422 | 2/4/2026 |
| 1.0.0-rc5 | 384 | 2/3/2026 |
| 1.0.0-rc4 | 413 | 2/3/2026 |
| 1.0.0-rc2 | 268 | 2/3/2026 |
| 1.0.0-alpha4 | 69 | 2/2/2026 |
Initial NuGet package release.
Signal extraction pipeline: embeddings, entities, knowledge graphs, ranking, LLM routing.