FinAI: An Investment Platform Built to Make Users Money — Not Sell Subscriptions

Most retail investment products make money whether you do or not. Subscription fees, order-flow rebates, and "premium tip" upsells all pay out regardless of whether the user's portfolio goes up or down. The incentive is engagement, not outcomes. FinAI started from the opposite premise: the goal is not to sell subscriptions — it is to make the user money. Every architectural decision in the platform falls out of that one sentence.

FinAI is a user-aligned investment decision-support platform for Turkish investors, built on Laravel. It combines a Turkish financial-literacy academy, realistic backtesting, adversarially-verified AI signals, and a tamper-proof transparency ledger. It is deliberately positioned — in this pre-license phase — as an investment research and decision-support tool: education, backtesting, paper trading, and portfolio monitoring. Personalised investment advice requires a CMB (SPK) license in Turkey, and every surface of the product carries a clear "this is not investment advice" disclaimer.

Glass-Box Signals on a Hash Chain

The most common failure mode of "AI stock picker" products is that you cannot audit them. A signal appears, it's right or wrong, and the track record is whatever the vendor chooses to show you. There is nothing stopping a provider from quietly deleting the losers.

FinAI makes that impossible by design. Every published signal is written to a SHA-256 hash chain — each record includes the hash of the previous one, so the entire history is cryptographically linked. If any past signal is edited or deleted after the fact, the chain breaks and verification fails. The ImmutableLogService handles this, and the public transparency page lets anyone re-run the verification and see the scoreboard of every call the system has made — wins and losses alike. Our own tests include explicit tamper-detection cases that mutate a historical record and assert the chain reports itself as broken.

This is what we mean by a glass box rather than a black box: the reasoning, the track record, and the integrity of the record are all open to inspection.

Adversarial AI Verification: Every Buy Signal Faces Three Skeptics

A single AI model generating buy signals is a liability. Language models are confident even when wrong, and a lone "generator" agent has every reason to produce signals because producing signals is its job. So FinAI never lets a generator's output reach the user unchallenged.

Every buy signal is run through an AdversarialValidator: three independent skeptic lenses — risk, liquidity, and macro — each tasked with trying to refute the signal rather than confirm it. If two or more of the three lenses reject it, the signal is never shown to the user. This is the same adversarial-verification pattern we use in our own engineering reviews, applied to financial recommendations: independent perspectives, biased toward refutation, majority-kill.

Crucially, the AI never bypasses the safety layers. The optional Gemini integration only produces candidates; those candidates still pass through the adversarial panel, the rule filter, and the immutable ledger before anything is published. AI widens the funnel of ideas — it does not get to skip the checks.

Non-Negotiable Circuit Breakers

On top of the adversarial panel sits a RuleFilter — a set of hard limits that no signal can override regardless of how confident the model is. A confidence threshold below which nothing is published. A volatility circuit breaker that suppresses signals when the market regime is too turbulent to trust. A drawdown stop that halts new recommendations when losses cross a defined line. These are deliberately not "AI judgment calls" — they are mechanical, non-negotiable boundaries, because the entire point is to constrain the model, not to trust it.

Backtests That Include the Taxes Nobody Else Models

A backtest that ignores costs is marketing, not research. Plenty of platforms show beautiful equity curves that quietly assume zero commission, zero slippage, and zero tax. Those curves do not survive contact with a real brokerage account.

FinAI's BacktestEngine models commission and slippage on every simulated trade — and, importantly for the Turkish market, it applies withholding tax (stopaj) via a dedicated TurkishTaxCalculator. The output is the set of metrics that actually matter: CAGR, Sharpe ratio, maximum drawdown, and win rate, computed after costs. When tax law changes, the rates live in a config file — config/finai.php — so updating them never requires touching engine code. The backtest lab ships with multiple strategies out of the box (buy-and-hold, SMA crossover, RSI, momentum) so users can compare approaches on the same realistic, after-tax footing.

A Portfolio Guardian, Not a Broker Integration

FinAI does not connect to your brokerage account, and that is a feature, not a limitation. Portfolio data is entered manually. This is deliberate data minimisation: there is no broker credential to leak, no read access to a live trading account, and no dependency on a third-party integration that could break or be compromised.

What the platform does with that data is watch for the risks investors tend to ignore. The PortfolioAnalyzer computes concentration via the Herfindahl-Hirschman Index (HHI) and flags when a portfolio has drifted away from the user's stated risk profile. The job of the guardian is not to chase returns — it is to catch the silent accumulation of risk that turns a good year into a ruinous one.

Education First

Before any of the signal machinery, FinAI puts a Turkish financial-literacy academy. The reasoning is simple: a user who understands what a drawdown is, why diversification matters, and what a Sharpe ratio actually measures is a user who can evaluate the platform's output critically rather than following it blindly. An informed user is the only kind of user a make-the-user-money product can honestly serve.

Where AI Fits — and Where It Doesn't

FinAI integrates Google Gemini (default model gemini-2.5-flash) at exactly two swap points: the signal reasoner and the autonomous blog writer. Both follow the same two principles.

First, the safety layers are immutable. Gemini only proposes candidates; the adversarial panel, rule filter, and hash-chain ledger still run on everything. Second, graceful degradation is mandatory. The GeminiClient never throws — on a missing key, a network error, or malformed output, it falls back to a deterministic implementation (RuleBasedReasoner for signals, a template generator for blog posts). The application runs fully offline, with no API key, and simply uses the deterministic path. AI is an enhancement layer, never a single point of failure.

The blog itself is autonomous: a weekly content calendar assigns each day a theme — markets, education, strategy, transparency, tax, portfolio, weekly review — and a daily scheduled command grounds each post in real platform data (actual backtest metrics, the live signal scoreboard, worked tax examples) before optionally enriching it with Gemini. Ground first, embellish second — never the other way around.

The Roadmap Is a Compliance Path

FinAI's roadmap is structured around Turkish regulation rather than feature lust. Phase 1 — the current platform — is the pre-license surface: education, backtesting, paper trading, and portfolio monitoring. Phase 2 adds general (non-personalised) investment commentary under the relevant CMB framework and an AI portfolio guardian. Phase 3, contingent on a CMB license, introduces genuinely personalised advice and performance-linked pricing — the point at which the platform's incentives and the user's are formally aligned. Phase 4 expands across asset classes, from BIST equities to TEFAS funds and beyond.

That ordering is intentional. We would rather ship a narrower product that is honest about what it is than overreach into regulated territory and undermine the trust the whole platform depends on.

Why This Matters

The hard part of building an investment product is not the indicators or the charts — those are commodity. The hard part is resisting every incentive to optimise for engagement, retention, and upsell at the expense of the user's actual returns. FinAI's answer is structural: align the goal (make the user money), make the track record impossible to fake (hash chain), make the AI prove itself against skeptics (adversarial panel), put hard limits the model cannot override (rule filter), and model the costs everyone else hides (after-tax backtests).

It is a decision-support tool, not a crystal ball, and not investment advice. But it is built so that when it is right, you can verify it — and when it is wrong, it cannot pretend otherwise.