embabel/dice
Knowledge graph construction and reasoning library
What's novel
Knowledge graph construction and reasoning library
Code Analysis
9 files read · 4 roundsA Kotlin/Spring library that extracts confidence-weighted propositions from text via LLM, resolves entities through escalating strategies, and projects the proposition store to multiple typed views (vector search, Neo4j graph, Prolog inference, agent memory) with a full lifecycle management system i
Strengths
Exceptionally well-architected with clean separation of concerns (pipeline doesn't persist, projectors compose rather than duplicate, SPI-based policy injection for trust/conflict/authority). The revision system's three-tier fast path (canonical text → vector similarity → LLM classification) with batch processing and optional cheaper LLM for classification shows genuine engineering sophistication. The dual-clock design (contentRevised vs metadataRevised) and idempotent evidence absorption are th
Weaknesses
The Prolog projection is relatively thin — it maps relationships to facts but the actual inference engine integration (tuProlog) and rule-based reasoning depth wasn't fully verifiable. The project is tightly coupled to the Embabel agent framework (provided-scope dependencies), making it less standalone. Some advanced features (dream loop, collector, consolidation passes) are marked experimental and their full implementation depth is uncertain.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
163
Contributors
5
Files
199
Active weeks
10
Repository
Language
Kotlin
Stars
24
Forks
6
License
Apache-2.0