State model
Short-lived turn state references an immutable transcript and context manifest. It does not copy source records or create durable customer memory by default.
A conversation can move gradually from factual product information into a personalised judgement without a clean linguistic boundary.
The controller evaluates each proposed response against customer-specific inputs, comparative language, implied recommendation and foreseeable consequence before selecting explain, calculate, clarify or specialist-review routes.
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.
Typed commands, state changes and receipts cross an event spine without surrendering domain ownership.
Permitted workThe router selects a certified route and context envelope. It does not inherit the selected specialist's business authority.
Consistency rulePreserve atomic intents, active journey, current custodian and route version across every turn and modality change.
Hard boundaryThe model is not a system of record, identity provider, policy authority or proof that an external effect occurred.
A customer or service operator needs a grounded answer from live state and controlled knowledge, without a direct business effect.
Personal circumstances may support an approved calculation but cannot be used to rank products or infer suitability outside an authorised journey.
A deterministic outer workflow contains model-led work inside typed, observable calls. Dashed messages remain proposals until policy or a human grants authority.
Short-lived turn state references an immutable transcript and context manifest. It does not copy source records or create durable customer memory by default.
Use direct read APIs for stable same-domain calls. Use a mediated gateway when the call crosses trust boundaries, needs shared throttling, or exposes reusable capabilities.
Bind all facts to a snapshot or as-of time. A partial source set remains partial; the system never substitutes a plausible zero or stale value.
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 atomic intents, active journey, current custodian and route version across every turn and modality change.
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.
peak_read_qps = concurrent_sessions x turns_per_minute x source_fanout / 60p95_turn_latency = max(required_source_reads) + policy + model + validationcontext_cost = admitted_input_tokens x routed_turnsFilter candidates and context before model classification so unauthorised capability descriptions or customer facts do not enter the prompt.
A design is production-ready only when teams can prove what happened, recover it and change it safely.
multi-intent and overlap test
low-confidence clarification
loop and maximum-hop test
unsupported-route and queue-rejection recovery
Turn history, admitted personal facts, response classification, boundary tests, prohibited alternatives, route decision and customer-facing wording.
An appropriately authorised adviser owns personalised recommendation or suitability decisions.
Begin with one read-only journey and explicit fallbacks. Add personalisation only after freshness, source precedence and context-minimisation evidence hold.
Journey, routing, conduct, channel and assisted-service owners define route and handoff policy.