FinWise AI – Institutional Wealth Advisory & Decision Support System (Pin & Paper)

Financial Telemetry Questionnaire

Working Memory for Mamdani FIS & Safety Priority Rules

Test Personas:

Full Name User Fact

Age (Years) Fuzzy Variable [18-80]

Gross Monthly Income (INR) Cash Inflow

₹

Essential Living Expenses (INR) 50% Needs Baseline

₹

Liquid Savings & Bank Balance (INR) Emergency Buffer

₹

Total Outstanding Debt (INR) Liabilities

₹

Monthly Loan EMIs (INR) Debt Ratio

₹

Highest Debt Interest Rate (% APR) Debt Avalanche

Investment Horizon (Years) Time Target

Loss Tolerance Subjective Attitude

Income Stability Earnings Security

Primary Financial Goal Objective

Fuzzy Centroid Score

52.4

MODERATE RISK (0-100)

Monthly Investable Surplus

₹28,000

After living expenses & EMIs

Debt-to-Income Ratio

7.7%

HEALTHY DEBT (<20%)

Emergency Reserve Coverage

1.4 Mo

DEFICIT (TARGET: 6 MO)

Recommended Asset Allocation

Mamdani & Rule Constrained

50/30/20 Budget Breakdown

Needs vs Wants vs Surplus

Forward Chaining Rule Execution Trace Log

Priority Execution Queue (P = 100 ... 70)

Priority	Rule ID	Name	Status	Condition Fired & Explanatory Trace

Mamdani Fuzzy Rule Base Firing Analysis (22 Rules)

TRIGGERED: -- / 22

Firing Strength Computation (T-Norm Min): α = min(μ_Age, μ_Income, μ_LossTol, μ_Horizon) × Weight. Rules with α > 0.000 are triggered and clip the output fuzzy set; rules with α = 0.000 are non-firing / inactive.

Rule ID	Antecedents (IF Conditions)	Consequent	Firing Strength (α)	Firing Status

Mamdani Fuzzy Inference Process (Matplotlib Visualizer)

CORE AI THEORY

ALGORITHMIC ARCHITECTURE 4-Stage Continuous Mamdani FIS Pipeline

Mathematical Invariant: ∑ Allocation = 100.0% · Zero Cliff Discontinuities

STAGE 1 μ_A(x) ∈ [0, 1]

Fuzzification

Maps crisp inputs (Age, Horizon, Loss Tolerance, Stability) to linguistic truth degrees via triangular & trapezoidal curves.

STAGE 2 α_i = min(μ_j)

T-Norm Rule Firing

Evaluates 9 deterministic rules simultaneously. Multi-antecedent conjunction uses Gödel Min T-Norm to compute firing truth strength α.

STAGE 3 μ_Agg(y) = max_i

Mamdani Implication

Clips output consequents (Conservative, Moderate, Aggressive) at rule firing heights and aggregates via S-Norm maximum operator.

STAGE 4 R* = ∫y·μ(y) / ∫μ(y)

Centroid Defuzzification

Calculates Center of Gravity over the aggregated envelope to yield a crisp risk score R* ∈ [0, 100], mapped smoothly to asset equity.

Mamdani Fuzzy Inference Matplotlib Process

Grounded Financial Advisory Report

Live Macro Citations (DuckDuckGo Grounding):

Hello! I am your FinWise Fiduciary Assistant. I have evaluated your financial inputs using Mamdani Fuzzy Logic and verified capital safety rules.

Ask me follow-up questions about your emergency reserve shortfall, debt payoff strategy, or asset allocation.

How We Evaluated FinWise AI: Making Sure the Advice is Safe & Correct

Test 1: The 'No Crazy Jumps' Test

Question: As someone gets older or takes on debt, does their risk advice jump erratically?

Scenarios Tested 1,296 Grid Points

Test Coverage (Zero Blank Spots) 100.0%

Age Smoothness (Older = Safer) 100.0%

Horizon Smoothness (Longer = Growth) 95.8%

Overall Smooth Progression 98.6%

Cliff-Edge Free Check PASS (Smooth)

Test 2: The 'Math Always Adds Up' Test

Question: Can the AI ever recommend >100% of money or ignore emergency cash?

Extreme Stress Profiles Tested 50,000 Monte Carlo

Asset Allocation Sums to 100% 100.0% Exact

50/30/20 Budget Balanced 100.0% Exact

Zero Negative Money Allocations 100.0% Pass

Emergency Fund Prioritized First 100.0% Enforced

Zero Rule Conflicts or Crashes 100.0% Conflict-Free

Test 3: The 'Certified Planner' Test

Question: Does FinWise give advice that a human Certified Financial Planner (CFP) agrees with?

Real-World Case Studies Tested 150 CFP Benchmarks

Agreement with Human Planners 97.33% Match

Statistical Correlation (Pearson r) 0.552 (Positive)

Average Difference vs Benchmark 13.95% (Prudent)

Senior Capital Protection (Age 60+) 100.0% Safe

Real-Time Web Search Verification Active (RBI / G-Sec)

Test 4: 17-Yr Market Backtest (2008–24)

Question: How did the AI's allocation actually perform through real historical crashes (2008 Lehman, 2020 COVID)?

Market Data (Authentic Returns) NSE Nifty 50, CRISIL, Gold

FinWise Moderate 17-Yr Wealth ₹5,19,730 (10.18% CAGR)

Risk-Adjusted Sharpe Ratio 0.302 (vs 0.123 Nifty 50)

Max Drawdown (Tail Risk Cut) -17.4% (vs -51.8% Nifty)

2008 Lehman GFC Crash Result -17.4% (Gold +29.5% Offset)

Sortino Downside Ratio 0.559 (Superior Defense)

EMPIRICAL EVIDENCE 17-Year Historical Market Backtesting Visual Proof (2008 - 2024)

Authentic Series: Nifty 50 TRI · CRISIL Bond Index · MCX Gold · RBI Repo

17-Year Historical Market Backtesting Performance

Frequently Asked Evaluation Questions (Demystified for Evaluators)

Q1: Is FinWise AI a Machine Learning (ML) model?

No. Standard ML models (neural nets, regression) are black boxes that can hallucinate numbers and cannot guarantee that asset allocations sum to 100%. FinWise AI is a Knowledge-Based Expert System that pairs continuous Mamdani Fuzzy Logic with deterministic safety rules. Every single rupee is mathematically accounted for.

Q2: Where did the 50,000 entries and 150 case studies come from?

We used automated Monte Carlo simulation to generate 50,000 extreme financial edge cases (zero savings, massive debt, retirement cliffhangers) to stress-test rule safety. The 150 benchmark case studies are official curriculum client archetypes used in Certified Financial Planner (CFP) training.

Q3: How do we know the recommended advice is correct?

First, by mathematical verification: portfolio allocations strictly total 100%, and emergency cash is always funded before buying stocks. Second, by fiduciary concordance: in 97.33% of case studies, FinWise chose the exact same conservative/moderate/aggressive allocation as certified human financial planners.

Q4: Why does FinWise outperform simple rules of thumb?

Naive rules like 100 - Age only look at age. FinWise AI's fuzzy logic simultaneously weighs loss tolerance, income stability, and goal horizons. For a risk-averse 25-year-old, it safely reduces volatile stocks; for a stable 45-year-old, it allows wealth compounding while protecting emergency reserves.

Q5: Did you test this with actual historic market crashes (Backtesting)?

Yes, comprehensively. We backtested FinWise allocations across 17 years (2008–2024) of authentic Indian market returns (NSE Nifty 50 TRI, CRISIL Composite Bond Index, Domestic Gold, and RBI Repo). During the 2008 Lehman GFC crash when 100% equity collapsed by -51.80%, FinWise Moderate limited losses to just -17.44% because of non-correlated gold (+29.5%) and sovereign debt (+10.2%). Over the full 17 years, FinWise Moderate delivered a Sharpe Ratio of 0.302 (vs 0.123 for Nifty 50), achieving superior final wealth (₹5.19 Lakh vs ₹4.86 Lakh) with half the volatility.