VeritasGraph Knowledge Graph Architecture
SpaceLizit grounds legal intelligence and automated petition drafting using VeritasGraph, an open-source GraphRAG framework developed by Bibin Prathap.
Open-Source Attribution & Specifications
Core Value: Eliminating Probabilistic Guessing in Legal AI
Traditional vector retrieval converts documents into unstructured dense embeddings and retrieves chunks based on statistical similarity. While useful for casual conversational search, vector search breaks down in law because it cannot follow relational chains across multiple exhibits and statutes.
VeritasGraph replaces probabilistic chunk matching with deterministic multi-hop graph traversals. Every case entity—petitioner, beneficiary, educational degree, Department of Labor prevailing wage level, and USCIS form field—is structured as an explicit vertex connected by typed legal relationships.
How VeritasGraph Powers SpaceLizit
1. **Multi-Hop Traversal**: Resolves complex legal queries across 5+ relationship layers in under 50 milliseconds.
2. **Statutory Pinpoint Grounding**: Binds every generated paragraph directly to 8 CFR § 214.2, 20 CFR § 655.760, or USCIS Policy Manual precedents.
3. **10,000+ Pre-Filing Denial Rules**: Traverses petition data against historical USCIS Request for Evidence (RFE) heuristics prior to submission.
4. **Data Sovereignty**: Runs natively in single-tenant customer VPCs with absolute zero model training on law firm work product.