System Overview

The engine is a Cargo workspace with 7 crates organized as a unidirectional pipeline.

Crates

Crate	Source files	Tests	Purpose
nis2-ingestion	chunk.rs, parser.rs, types.rs	3	JSON parser, semantic chunking by legal hierarchy
nis2-vectordb	embed.rs, lance.rs, store.rs	1	LanceDB storage, BGE-Small embeddings, HNSW search
nis2-rules	engine.rs, obligations.rs, schema.rs, validation.rs, adapters.rs	20	Boolean decision trees, obligation catalog, JSON Schema validation
nis2-mcp-server	protocol.rs, tools.rs, server.rs	13	MCP JSON-RPC 2.0 server over stdio
nis2-slm	prompt.rs, grammar.rs, inference.rs, report.rs	18	Prompt construction, GBNF grammar, template report generation
nis2-api	handlers.rs, routes.rs	6	Axum REST API, benchmark binary
nis2-chat	engine.rs, cache.rs, dataset*.rs	20	Semantic cache, BM25 search, session tracking, multilingual Q&A

Data flow

data/sources/*.json -> ingestion (parse+chunk) -> vectordb (embed+index)
                                                        |
CompanyProfile -> rules (evaluate) -> ComplianceStatus -> slm (report)
                       |                    |
                  mcp-server            api-server
                  (stdio)              (HTTP :8080)

Binaries

Binary	Crate	Purpose
indexer	vectordb	Populate LanceDB from JSON sources
search	vectordb	Semantic search over indexed provisions
demo	rules	Print ComplianceStatus JSON
report	slm	Generate Italian compliance report
mcp-server	mcp-server	MCP JSON-RPC server
api-server	api	Axum HTTP server
bench	api	Benchmark suite
chat	chat	Interactive CLI chatbot