System Architecture

Deterministic computation meets AI narration

Every number is traceable, auditable, and reproducible. The engine handles the maths. The LLM handles the narrative.

"Give AI facts to explain, not problems to solve. The engine handles the maths. The LLM handles the narrative. Every number is traceable, auditable, and reproducible."
— Architecture principle
01 — User Inputs

Three categories of inputs define a retirement scenario. Everything the simulation needs comes from the user — no hidden assumptions.

📊

Portfolio Weights

Asset class allocations across equities, bonds, real estate, commodities, and cash.

💰

Cash Flows

Income streams, pension, CPF, SRS, annuities, planned withdrawals & expenses.

⚙️

Parameters

Retirement age, horizon, inflation, tax rules, risk tolerance, asset correlations, expected returns.

↓
02 — Computation Engine

All calculations are deterministic and reproducible. Given the same inputs, the engine produces identical outputs every time — no AI involved at this stage.

🔄

Monte Carlo Simulation

600,000 scenarios with correlated asset returns, inflation modelling, and cash flow projections.

Deterministic
🔥

What-If Heat Maps

Two-dimensional optimisation across allocation × withdrawal rates showing full solution space.

Deterministic
↓
03 — Pre-Calculated Results

The engine packages results into a structured data object. This is the single source of truth — success probabilities, percentile outcomes, risk metrics, cash flow tables, and chart data. No opinions. Only facts.

📋

PublishMyPlan

Structured data package: success probabilities, percentile outcomes, risk metrics, cash flow tables, charts.

Facts, not opinions
↑ Deterministic & Auditable · AI Narration ↓
04 — AI Narrative Layer

Pre-calculated results are passed to the LLM with strict guardrails. The prompt engineering enforces a clear boundary: explain these facts — do not infer, do not hallucinate, do not add data.

📝

Prompt Engineering

Strict instructions: explain these facts, do not infer, do not hallucinate, do not add data.

Guardrailed
🤖

LLM API

Claude Opus receives pre-calculated results + prompt. Returns narrative report with enhanced explanations.

AI-Powered
↓
05 — Client Deliverable

The final report combines deterministic charts, tables, and data with AI-generated narrative insights — institutional quality, retail accessible.

📄

AI-Enhanced Retirement Report

Professional wealth management report combining deterministic charts, tables & data with AI-generated narrative insights — institutional quality, retail accessible.

Computation + AI Narration