How the engine thinks

Not a single forecast — three competing career-and-life strategies, each projected to the same goal year with the same engine. Every path is scored on endgame wealth, on-track probability, and cross-border readiness, then ranked side by side.

Each scenario carries its own compensation trajectory, equity grants, relocation timing, and tax exposure. The engine runs identical math across all three so the comparison is apples-to-apples — the only variables are the life choices.

Every plan runs 5,000 simulations with antithetic variates and a seeded PRNG — reproducible results, not random noise. The headline is a probability (the odds of clearing the goal) with a P10/P50/P90 cone, not a single tidy line pretending the future is certain.

Antithetic variates pair each random draw with its mirror, cutting variance in half for the same number of runs. The seeded generator means you get the same result on every page load — until you change an assumption.

Every input is modeled worst-case to best-case — conservative and aggressive returns, base and stretch career arcs, weak and strong rupee. You see the honest downside and the full upside, never just the rosy midpoint.

Uncertain upside (like rental income on a property you haven't bought) goes into the ceiling only, never the floor. The floor is what you can count on; the ceiling is what's possible if things break your way.

Drag any assumption — equity return, FX drift, inflation, savings rate — and the full projection recomputes instantly in your browser. No server round-trip, no loading spinner. The model isn't a black box; it's a thing you can poke.

The engine is deterministic and client-side: the same inputs always produce the same outputs. Changing one lever reruns the entire Monte Carlo + scenario stack in milliseconds, so you feel the tradeoff immediately.

Checking, savings, credit cards, and retirement accounts sync via Plaid. Net worth updates every time you open the app — no manual entry, no stale spreadsheet.

Your self-directed brokerage is itemized holding by holding, each repriced from live market quotes. Unrealized gain/loss, today's move, and the full position table — mirroring your broker, but wired into the projection engine.

Amazon RSUs are tracked grant by grant with the real vesting schedule, repriced at today's AMZN. Vested shares feed the brokerage; unvested shares feed the scenario projections. A pending Datadog offer is modeled as "projected" with a dashed border and a badge — it never masquerades as settled wealth.

Each vest date, share count, and grant origin is explicit. The engine prices unvested shares at today's stock price and projects them forward as part of the career scenario, so the 10-year trajectory includes equity you haven't received yet.

Your US home value comes from RentCast (Zillow-grade AVM), and the engine calculates amortized mortgage payoff, closing costs, and net proceeds — timed to your planned move year. The sale proceeds flow into the cross-border projection automatically.

The USD/INR rate and gold spot price are pulled live, not hardcoded. Your Indian assets and gold holdings are valued at today's rate, and rupee depreciation is baked into the long-term projection — not as a guess, but as a modeled drift with a range.

The engine scans your full 6-year transaction ledger to identify recurring charges — subscriptions, rent, insurance, loan payments — by frequency and merchant pattern. It beats Plaid's built-in recurring detection because it works from the raw history, not a third-party model.

Every transaction lands in one of ten buckets — five mandatory (housing, food, transport, health, debt), five discretionary (personal, travel, shopping, entertainment, subscriptions). Each bucket is labeled cuttable or non-negotiable, so the runway calculator knows exactly where the slack is.

The taxonomy is designed for a specific purpose: answering "how long can I survive if I cut everything optional?" The engine uses the mandatory/discretionary split to stress-test your runway under different austerity scenarios.

Claude reads your categorized transactions and writes plain-English observations — spending anomalies, trend breaks, seasonal patterns — that surface as insight cards on the dashboard. Not canned alerts, but contextual analysis of your actual numbers.

Ask anything about your finances in plain English. The engine interprets the question, pulls the relevant data, and answers with numbers and context — no menu-diving, no query syntax.

Most apps project runway from a budget you set. This engine uses your real trailing 12-month spending — after excluding one-time outliers — to calculate how many months of cash you have. The number is honest because the input is real.

The engine separates one-time spikes (a car down payment, a medical bill) from your steady-state burn rate. Runway is based on the steady state, so a single large purchase doesn't panic the model.

Three tiers of stress: current burn rate, belt-tightened (mandatory-only spending from the 10-bucket split), and worst-case (job loss + mandatory only). Each scenario shows months of runway and the cash-out date, so you know exactly how much margin you have.

Compensation assumptions aren't guesses — they're sourced from Glassdoor, Levels.fyi, and industry surveys, with the source cited next to every number. When the engine says "Senior Product Designer at Google in Hyderabad: ₹95–99L," that's a researched range, not a vibes estimate.

US dollar and Indian rupee assets are tracked natively — bank accounts, real estate, gold — each in its local currency, converted at the live FX rate for the combined view. No manual currency juggling.

The projection doesn't assume a fixed exchange rate. It models rupee depreciation as a range (floor and ceiling drift), so your ₹-denominated goal is stress-tested against currency headwinds. Indian assets that are priced in rupees are immune to this depreciation — the engine knows the difference.

The Bluffdale property sale is modeled with a live valuation, amortized payoff schedule, and closing costs — timed to the year you plan to relocate. The net proceeds inject into the projection at exactly the right point in the timeline.

The ₹70 Cr target includes ₹ trusts for each child, carved out as a separate line in the goal breakdown. The engine tracks "after trusts" wealth separately, so you always know what's yours versus what's earmarked.

All computation — projections, Monte Carlo, scenario scoring — runs locally on your device. There is no cloud backend storing your financial data. The app reads your data files and computes everything in-process.

Plaid connects to your banks with your explicit authorization, and the credentials live with Plaid's infrastructure (never in the app). You can disconnect any account at any time, and the app continues to work with cached data.

When the engine calls Claude for insights or question-answering, it sends the minimum context needed. No financial data is stored by the AI provider, no model is trained on your numbers, and every AI-generated insight is labeled as such.

The projection engine is pure math — given the same inputs, it produces the same outputs, every time. No opaque ML model deciding your future. Every assumption is visible, every number is traceable, and the methodology is documented right here.