State model
The source document remains immutable. Derived rules and fragments carry source anchors, compiler version, effective dates and approval state.
Bank employees need fast answers across policies, procedures and product rules whose authority, date and audience differ.
The assistant routes each question to effective, role-permitted sources and answers with claim-level citations, conflict disclosure and an explicit no-answer state.
Organisations, systems and operating conditions are intentionally anonymised and recomposed. The design demonstrates engineering and banking-domain reasoning; it does not represent a named client estate, vendor product or measured production result.
Five operating planes separate interaction, identity, decision control, authoritative state and operating evidence.
Permitted workThe system may summarise approved work information and suggest next steps. Employment, performance and access decisions remain authorised human processes.
Consistency rulePreserve authorship, decision status, supersession and source permissions; a meeting summary cannot create an approved decision.
Hard boundaryThe model is not a system of record, identity provider, policy authority or proof that an external effect occurred.
Documents, rules and product content must become effective, attributable and executable without allowing retrieval relevance to decide authority.
Retrieval entitlement is evaluated before search, and an answer cannot outrank the underlying policy source.
A deterministic outer workflow contains model-led work inside typed, observable calls. Dashed messages remain proposals until policy or a human grants authority.
The source document remains immutable. Derived rules and fragments carry source anchors, compiler version, effective dates and approval state.
Serving APIs expose rules and evidence separately. Retrieval returns candidates; an authority filter decides whether they may enter context.
A request resolves one effective rule set for its product, jurisdiction and date. Conflicting versions stop consequential conclusions.
These roles are deliberately vendor-neutral. Each can be independently owned, versioned and replaced.
Accepts a typed evidence request and returns the minimum permitted facts with source, event time, observation time, validity and exclusions.
Evaluates identity, purpose, capability, amount, risk tier and policy version; returns allow, deny, step-up or human-review with reasons.
Chooses an approved model route by task, risk, evidence quality, latency budget and cost ceiling; enforces structured outputs.
Serves owned, audience-qualified and effective-dated content; exposes supersession, withdrawal and dependency metadata.
Appends request, versions, policy result, model proposal, approval, action receipt, readback, correction and custody events under one correlation key.
Durable records carry provenance, authority, effect and custody without turning a transcript into an uncontrolled memory store.
Preserve authorship, decision status, supersession and source permissions; a meeting summary cannot create an approved decision.
The selected design is not universally superior. It is the safer fit for this boundary and failure cost.
Prefetch stable, purpose-safe facts; acquire volatile facts on demand against a time-qualified snapshot.
Load a broad customer profile at session start.
Cost acceptedThe selected design adds source calls and latency, but reduces stale data, excess exposure and accidental reuse.
Compile stable decision logic and retain retrieval for explanation and residual ambiguity.
Ask a model to interpret the source document for every request.
Cost acceptedRule compilation needs controlled change, but creates repeatable decisions, regression tests and clear exceptions.
Reserve larger models for residual reasoning after deterministic and smaller-model gates.
Send every request to the most capable available model.
Cost acceptedRouting adds evaluation work and operational complexity, but controls cost, latency and unnecessary data exposure.
Federate authoring while centralising lifecycle metadata, validation and serving rules.
Create one centrally authored knowledge corpus.
Cost acceptedFederation requires stronger contracts and owner discipline, but preserves domain accountability and release velocity.
Use append-only events plus a rebuildable current-state projection.
Overwrite the case row with its latest status.
Cost acceptedReplay and storage are more complex, but point-in-time reconstruction and correction lineage remain possible.
Retries are bounded by knowledge of business effect; unknown outcome remains visible, owned and independently reconciled.
Actual thresholds belong to accountable service owners. The design exposes the equations and observables that those owners must baseline.
compile_work = changed_sources x affected_rules x dependent_routesserving_qps = knowledge_requests x candidate_fragmentsrevalidation_scope = changed_dependencies x linked_test_setsApply source permissions at retrieval time and prevent private messages, sensitive HR data or draft content from crossing purpose boundaries.
A design is production-ready only when teams can prove what happened, recover it and change it safely.
source-permission trimming
decision versus discussion classification
sensitive-data exclusion
feedback and correction lineage
Question purpose, entitlement result, retrieved passages, source versions, conflict handling and feedback outcome.
Policy owners resolve conflicts; bank employees remain accountable for the business decision.
Start with high-consequence rules and stable documents. Expand automation only where extraction corrections and owner decisions show repeatable structure.
Business, HR, knowledge, security and information-management owners define permitted uses.