How Prism Works

An AI-powered portfolio analysis engine built for Canadian self-directed investors. Deterministic math first, AI narration second.

Design Principles

🧮

Deterministic First

The math engine computes every metric before AI touches anything. Portfolio value, allocation drift, concentration risk, tax-loss candidates — all calculated with pure arithmetic.

🧠

AI Explains, Never Invents

GPT-4o-mini receives pre-calculated numbers and writes a narrative. It cannot invent statistics. If no API key exists, a deterministic fallback generates the explanation.

🎯

One Clear Recommendation

Instead of overwhelming you with options, the engine scores 7 possible actions and surfaces the single highest-value move. This prevents decision paralysis.

🍁

Canadian-Specific

TFSA and RRSP contribution room scoring, Canadian marginal tax rates for tax-loss harvesting, and Canadian ETF return assumptions are baked into every calculation.

The Engine Pipeline

Three pure functions process your portfolio data in sequence.

computeMetrics()

Calculates total value, cost basis, unrealized G/L, asset-class breakdown, allocation drift, weighted annual return, concentration risk, positions in loss, savings rate, and emergency fund coverage.

scoreActions()

Evaluates 7 possible actions (Deploy Cash, TFSA Optimize, RRSP Optimize, Rebalance, Tax-Loss Harvest, Reduce Concentration, Add to Position) and scores each 0–100 based on expected financial impact and time-sensitivity.

projectScenarios()

Projects portfolio growth over your goal timeline under Base, Recession, and Bull market conditions. Each scenario outputs a goal probability and year-by-year portfolio values.

Form Input→computeMetrics()→scoreActions()→projectScenarios()→AI Narrative→Dashboard

Action Scoring

Seven possible actions are evaluated every analysis.

ActionTriggerUrgency

Deploy CashExcess cash above target allocationHigh if >10% excess

TFSA OptimizeUnused TFSA room > $1KHigh if >$20K room

RRSP OptimizeUnused RRSP room > $5KMedium

RebalanceAllocation drift > 5%High if drift >15%

Tax-Loss HarvestPosition down > 8%High if savings >$2K

Reduce ConcentrationHolding > 25% of portfolioHigh if >40%

Add to PositionGain >15%, weight <15%, above-avg returnLow

The AI Layer

GPT-4o-mini receives the pre-calculated outputs from the engine and generates:

A punchy headline
2–3 paragraph explanation
3 key numbers from the action details
A confidence score (constrained to ± a small range of the engine score)

Strict rules: The AI cannot invent numbers, must stay within confidence bounds, and always includes a disclaimer.

If no OpenAI API key is configured, a deterministic mock generates the narrative using the same data.

Tech Stack

Next.js 14

App Router, server-side API routes

TypeScript

End-to-end type safety

React 18

Client-side interactivity

Tailwind CSS

Utility-first styling

GPT-4o-mini

AI narrative generation

Vercel

Edge deployment, zero config

What's Next

Future development roadmap

✓ PDF / CSV export

Client-side PDF and CSV report generation

✓ Custom scenarios

User-defined return and shock parameters

✓ Holdings breakdown

Sortable per-holding gain/loss and weight table

✓ What-If Explorer

Real-time parameter tweaking with delta comparison

✓ Monte Carlo analysis

500-run fan chart with percentile bands

✓ Rebalancing worksheet

Concrete buy/sell trade list for target allocation

✓ Contribution gap calculator

Monthly shortfall/surplus to reach goal

Real-time pricing

Market data API integration (Alpha Vantage, etc.)

Portfolio persistence

Database + auth (Postgres / Supabase)

Brokerage account import

Open Banking API integration

Backtesting

Historical return data integration