Fintech

The ladder

Three products, in this order, for a reason.

Each step depends on the one before it. You can't advise on a book without a trustworthy read of the market, and you can't plan a life without knowing how the book behaves.

Market Intelligence

QuantVantage by Nylivo — what's happening in the market, cited. Live in production.

Portfolio & AI Advisor

What it means for your positions, and what to change about them. In design.

Financial Planning

What it means for your goals over decades. Roadmap.

All three read and write one Brain, so each product inherits the ground truth the previous one established — the planning product will reason over the same positions and the same provenance the advisor already trusts.

Step 02 · workspace anatomy

Ingest. Graph. Advise. Grade.

How the Portfolio Management & AI Advisor workspace is built — the same four stages as every other workspace on the platform, where only the third is domain-specific. That's the thesis: the industry is configuration.

Ingest

The real book, not a model of one

Advice built on approximated holdings is worth nothing. Positions come from the brokerage connection with cost basis and lot detail intact, and options are valued from live marks rather than a theoretical price that flatters the book.

Positions & tax lotsEquities, options, and cash with acquisition dates and cost basis — lot-level, because tax consequences are lot-level.
Live marksCurrent pricing for every instrument, with provenance. A position whose mark is missing is excluded from a statistic, never zero-filled into it.
Account structureRegistration type, tax treatment, and restrictions — what's actually permissible differs per account.
MandateObjectives, horizon, risk tolerance, liquidity needs, and explicit constraints. Recorded, versioned, and the basis for every suitability claim.
Market & regime contextVolatility regime and dealer positioning, read from the same market plane QuantVantage.today already runs on.

Graph

Risk hides in the second hop

The dangerous exposure is never on the position list. An ETF, a single stock, and a call spread can all load on one factor or one issuer — and a flat table will never tell you. Look-through concentration is a graph traversal.

EntitiesAccount, Position, Lot, Instrument, Issuer, Sector, Factor, Objective, Constraint, Proposal.
RelationshipsAccount holds Position → Lot · Position of Instrument · Instrument issued by Issuer · Issuer in Sector · Instrument loads on Factor · Proposal affects Position.
Look-throughFund and ETF holdings expanded to their underlying issuers, so “diversified” can be tested instead of assumed.
ProvenanceEvery number tagged live / store / cached / simulated — the difference between a real risk figure and a stale one is visible, not buried.

Advise

A proposal with its reasoning attached

The advisor's output is never a number on its own. It's a ranked, cited proposal — what to change, why, what it costs in tax, and what it does to risk — written so a professional can disagree with the reasoning rather than trust a score.

Risk cockpitEWMA volatility, VaR and CVaR (tail loss, not just the threshold), drawdown behaviour under stress.
Stress & scenariosNamed historical and hypothetical shocks applied to the real book, including the options leg.
Concentration & factorLook-through exposure by issuer, sector, and factor — the risk the position list conceals.
Tax-aware rebalancingLot-level SELL/ADD candidates with wash-sale awareness and harvesting opportunities. After-tax is the only rebalance that matters.
Options overlayBook-level greeks, an expiry desk, and assignment risk surfaced before expiration week, not during it.
Ranked proposalsCandidates scored and walk-forward backtested against a benchmark — never in-sample, because in-sample results are a marketing artifact.

Grade

Graded against doing nothing

The honest benchmark for advice isn't the market — it's the book you already had. Every proposal is scored against the do-nothing baseline at a fixed horizon, so “we added value” is a measured claim.

Do-nothing baselineEach proposal's realized outcome compared to leaving the position untouched. Advice that doesn't beat inaction is not advice.
Closed-only attributionOpen proposals are excluded from every published rate — no counting the ones still in flight.
Overrides are signalWhen a human edits or rejects a proposal, that's recorded and learned from. Being overruled is training data.
Bounded learningWeights adjust from closed outcomes within hard limits, so the model adapts without quietly drifting into a different strategy.

Regulatory posture

An AI advisor is a regulated claim.

Calling something an advisor invites scrutiny it has to survive. This is built as decision support for a licensed human — not an autonomous manager, and not a shortcut around the obligations that attach to advice.

A human owns the advice

The system proposes; a licensed professional (or the self-directed account holder) decides. No discretionary auto-trading — the confirm-first ticket is the same control already shipped in production.

Suitability is traceable

Every proposal resolves back to the recorded mandate — objectives, horizon, tolerance, constraints — as a versioned record, so the basis for a recommendation is reconstructable later.

Reproducible by construction

Inputs are retained with their as-of timestamps, so a recommendation can be re-derived months later from exactly the data that produced it. “The model changed” is not an acceptable answer to an examiner.

A complete audit trail

Who proposed, on what data, with what reasoning, approved by whom, and what actually changed — one immutable chain per action.

No fabricated performance

Track records come only from closed positions, benchmarked and dated. Backtests are labelled as backtests. Hypothetical is never presented as realized.

Model spend is accountable

Per-account LLM metering means the cost to serve advice is a known number — which matters when the economics have to survive a fee schedule.

This posture isn't theoretical: the platform's honesty gates, confirm-first trade ticket, and outcome ledger were all built and hardened in production before this product was scoped.