Context
A financial institution carrying BCBS 239 and IFRS 17 obligations has to be able to trace any reported figure back to its statement of record. That is the requirement. In practice the trace is usually reconstructed by hand, under deadline, by people who did not build the pipeline and are reading it for the first time.
The data exists. The relationships between the data do not, and every reporting cycle somebody rediscovers them from scratch.
Built
A semantic discovery and lineage engine that maps every transaction back to its statement of record.
The distinction that matters: the audit trail is generated as a product of the pipeline rather than as an archaeological exercise after the fact. The lineage is not a report about the system, it is what the system emits while it runs.
Stack
TypeScript, Next.js, Python and PostgreSQL, with semantic indexing and a multi-provider LLM abstraction.
Outcome
Built and maintained as a product rather than a one-off engagement. It applies directly to data management and governance practices that carry regulatory reporting obligations, which is a narrower claim than "financial services" and a more accurate one.