README
¶
11 · Unified Brain — two live databases, streaming CDC, and reasoning
A complex, multi-source CortexDB application. It merges two live databases into one privacy-safe knowledge graph, keeps both continuously in sync, and reasons over the result — with no LLM and no embedder (deterministic lexical retrieval + SPARQL + RDFS/SHACL).
Postgres (customers, orders) ─┐ connector desensitize ┐
MySQL (tickets, free-text) ┘ (signed plan) ├─▶ one CortexDB brain
│
continuous CDC: PG logical replication + MySQL binlog ─┘ (both stream concurrently)
│
reasoning: RDFS inference · SPARQL aggregates · SHACL validation
What it demonstrates
- Dual source → one graph. Customers/orders (Postgres) and support tickets
(MySQL) become one knowledge graph:
Customer ──placed──▶ OrderandCustomer ──raised──▶ Ticket. - Free-text PII redaction. Ticket bodies contain phone numbers and emails in
prose; the connector redacts them in place (
[REDACTED:phone],[REDACTED:email]) before indexing — including on the live CDC path. - Concurrent streaming CDC. A Postgres logical-replication watcher and a
MySQL binlog watcher run at the same time; a live
INSERTin either database lands in the brain within milliseconds. - RDFS inference.
Customer rdfs:subClassOf Partymaterializes a derivedrdf:type Partyfor every customer. - SPARQL aggregate. "Priority customers" =
COUNT(paid orders) >= 2viaGROUP BY … HAVING. - SHACL validation. A shape requires every
Customerto have acity; the demo seeds one customer without a city and the validator catches it.
Run it
docker run --rm -d --name cx_pg -e POSTGRES_PASSWORD=p -p 5432:5432 postgres:16 -c wal_level=logical
docker run --rm -d --name cx_my -e MYSQL_ROOT_PASSWORD=p -e MYSQL_DATABASE=test -p 3306:3306 mysql:8
go run ./examples/11_unified_brain \
-pg 'postgres://postgres:p@localhost:5432/postgres?sslmode=disable' \
-my 'root:p@tcp(localhost:3306)/test'
The program seeds both databases (including a publication for Postgres logical replication), so a fresh pair of containers is all you need.
Prerequisites
- Postgres with
wal_level=logical(the-c wal_level=logicalflag above). - MySQL with ROW binlog +
binlog_row_image=FULL(defaults inmysql:8).
No LLM, embedder, or API key — the whole pipeline is deterministic.
Documentation
¶
Overview ¶
Unified support brain — a complex, multi-source CortexDB application.
It unifies TWO live databases into one privacy-safe knowledge graph and keeps them continuously in sync, then reasons over the result:
Postgres (customers, orders) ─┐ connector desensitize ┐
MySQL (tickets, free-text) ┘ (signed plan) ├─▶ one CortexDB brain
│
continuous CDC: PG logical replication + MySQL binlog ─┘ (both stream concurrently)
│
reasoning: RDFS inference · SPARQL aggregates · SHACL validation
What it demonstrates beyond a basic RAG demo:
- DUAL SOURCE: customers/orders from Postgres + support tickets from MySQL, merged into a single knowledge graph (Customer ──placed──▶ Order, Customer ──raised──▶ Ticket).
- FREE-TEXT PII: ticket bodies contain phones/emails in prose; the connector redacts them in place before indexing.
- STREAMING CDC from BOTH at once: a Postgres logical-replication watcher and a MySQL binlog watcher run concurrently; live INSERT/UPDATE/DELETE in either database flow into the brain within milliseconds.
- REASONING: RDFS-lite inference materializes derived types; a SPARQL aggregate finds "priority" customers (>=2 paid orders); SHACL validates a data-quality constraint and catches a real gap.
No LLM or embedder required — deterministic lexical retrieval + SPARQL.
Prerequisites: a Postgres with wal_level=logical and a MySQL with ROW binlog (both are the defaults for `postgres:16 -c wal_level=logical` and `mysql:8`).
Usage:
go run ./examples/11_unified_brain \ -pg 'postgres://postgres:p@localhost:5432/postgres?sslmode=disable' \ -my 'root:p@tcp(localhost:3306)/test'
Source Files