RAG · Finance
RAG and document search for financial services
In a financial institution the source and date of an answer matter as much as the answer. Zero-Shot builds search systems over internal policy, procedure, regulatory and reporting archives: every query and answer is logged for audit, answers are tied to dated sources, and permission boundaries hold.
What we build
- 01Version-aware search across internal policy and procedure archives
- 02Date-indexed regulation and regulator correspondence
- 03An audit record of every query, the sources retrieved and the answer given
- 04Permission filters by department and role
- 05A diff view showing what changed between policy versions
How it runs
01
Scope and classification
Which document classes enter the system, and which never do, is written down first. Documents containing customer data are handled separately.
02
Dates and versions are mandatory
A policy document without an effective date is not indexed. Version differences are marked so the system can see them.
03
Audit trail
Who asked, which documents were retrieved, what was answered — all three stored together, readable by an auditor afterwards.
04
Measured with compliance
Correct answers in the evaluation set are labelled by the compliance team, so the accuracy claim comes from inside the institution.
What we measure
- Share of questions where the correct policy clause is the first result
- Number of times a superseded policy is presented as current
- Permission leak incidents (target: zero)
- Audit trail completeness — queries recorded without gaps
- Time to answer a compliance question
When this is the wrong fit
The system gives no investment advice, executes no transactions and makes no customer decisions. It is built for internal document access, not for individual client decisions based on personal financial data.
Common questions
What is recorded for audit?
For every query: who asked, the question itself, the source documents retrieved and the answer produced. The record is stored immutably, with a retention period set by your own policy.
Can it be deployed without data leaving the institution?
Yes. The system can run entirely on internal infrastructure with open-weight models. Which component runs where is written down before setup.
What happens to old answers when a policy changes?
Answers are not stored, sources are; each question is answered again from the currently valid document. Version diffs can also be shown, so 'what did this say last month' remains answerable.
Other work in this sector
Other services
What we've written on this
- Where to start with enterprise AIHow to choose a first AI project: which tasks qualify, in what order to proceed, and what to expect in the first eighty days.
- Why AI makes things up, and how to reduce itWhy language models state false things confidently. The real cause of hallucination, which countermeasures work, and which do not.
- Is it safe to send company data to AIWhat happens when company data reaches a language model: provider terms, training use, data protection law, cross-border transfer and self-hosting.
Last updated: 2026-09-05
Let's talk about this one
Tell us what you want to build and within two days we'll tell you who takes it and how we'd measure it.