Rules-first AI for sensitive decisions

A financial coaching product keeps credentials, calculations and permissions outside the language model, then uses AI only to explain bounded results.

A security boundary separating private data and deterministic rules from the language modelThe model can explain a decision without owning the numbers, credentials or action.Read-only data01Rules engine02Safe aggregate03AI explanation04
The model can explain a decision without owning the numbers, credentials or action.

01 / The field note

AI is most useful in sensitive systems when it is not the source of truth. A financial coaching architecture makes this concrete: deterministic code owns the calculation, and the model receives only the bounded result it needs to explain.

This note describes a personal product in development. It is not presented as a public financial service.

01

Put permissions before intelligence

Financial providers are connected through read-only access. The product has no payment initiation path, and the language model never receives bank credentials or provider tokens.

A separate security boundary prevents a convenient assistant feature from expanding into authority over money.

02

Keep numbers deterministic

Cash coverage, protected buffers, known bills and category limits are calculated by code with inspectable rules. Missing data reduces confidence instead of inviting the model to fill the gap.

The same inputs always produce the same numeric state, which makes testing and review possible.

03

Give the model a bounded job

The model receives sanitized aggregates, confidence and allowed coaching context. Its task is to explain trade-offs, ask for missing information or reframe the next action.

It cannot move money, change the ledger or redefine a safety rule.

04

Design for incomplete signals

Open-banking data can be delayed or rate-limited. The interface must show last-sync time, coverage and provisional status instead of presenting stale data as real time.

Safe AI depends as much on honest uncertainty as it does on prompt quality.

03 / Working principles

The reusable part

What to carry into the next system.

  1. 01

    Keep credentials and write authority outside the model boundary.

  2. 02

    Calculate sensitive values with deterministic, testable rules.

  3. 03

    Give the model sanitized aggregates and a narrow explanatory task.

  4. 04

    Display freshness, missing coverage and provisional confidence.

05 / Contact

AI · AWS · DevOps · WordPress · Software

Need this kind of decision in your system?

A discovery call is enough to map the constraint, identify the evidence still missing and decide on the smallest useful intervention.

Start with the problem

Tell me what needs to move.

A short description is enough. I will review it personally and reply with a useful next step.

By sending this enquiry, you confirm that you have read how the information is handled in Legal & privacy.