Module 09 · cognitive feature

Semantic ontology mapping

Legacy enterprise databases are chaotic, undocumented and cryptically named. The cognitive pipeline samples the schema, infers an entity mapping with a confidence score, and flags what a human still has to decide — which is what lets the agent answer 'what is our exposure to NMC Health?' against a 20-year-old Oracle instance.

$ python -m fde_toolkit ontology

Legacy schema

what the FDE finds on day one

TBL_CUST_99X_REV → Customer

no table comment in the source system

auto-accepted

A legal entity or individual the bank has onboarded.

Legacy column	Semantic attribute	Inferred meaning	Conf.
`CUST_ID_NO`	`customerIdentifierNumber`	customer identifier number (VARCHAR2(18)), e.g. CUST-9921	0.99
`CUST_NM_FULL`	`customerNameFull`pii	customer name full (VARCHAR2(120)), e.g. NMC Health PLC	0.99
`DOM_CTRY_CD`	`domicileCountryCode`	domicile country code (CHAR(2)), e.g. AE	0.99
`KYC_STS_FLG`	`knowYourCustomerStatusFlag`pii	know your customer status flag (CHAR(1)), e.g. V	0.99
`PEP_IND`	`politicallyExposedPersonIndicator`pii	politically exposed person indicator (CHAR(1)), e.g. N	0.99
`REV_DT`	`reviewedDate`	reviewed date (DATE), e.g. 2024-11-02	0.99

Enterprise knowledge graph

Customer—HOLDS→Account

Account—SENT→Transaction

Transaction—RECEIVED_BY→Account

Customer—HAS_EXPOSURE→CreditExposure

Customer—SUBJECT_TO→KycStatus

Natural language → Cypher

what the mapping buys you

MATCH (c:Customer {name: $name})-[:HAS_EXPOSURE]->(e:CreditExposure)
RETURN c.name, e.limitAmount, e.utilisedAmount,
       e.utilisedAmount / e.limitAmount AS utilisation

Without the ontology this question is unanswerable — no model knows thatEXP_LMT_TB.UTIL_AMT means utilised credit exposure. The mapping, not the model, is the deliverable.