An autonomous trading platform for European retail: a learning RL agent, sentiment from financial news, an institutional risk engine and a conversational AI copilot. Stocks and crypto, 24/7.
Product · working MVP
This isn't a mockup — all four services run live on a paper account at Alpaca and move to live after validation. The full set of capabilities:
A loop every few minutes, stocks and crypto, long and short, idempotent orders, multi-tenant.
A per-user model that learns from your trades. Versioning, promote/rollback, OOS gating.
Our own conversational assistant with access to your live account. Answers anything about trading and takes actions on your command.
Mood from hundreds of financial news items per minute + a news analyst (catalysts, risks).
Half-Kelly sizing, trailing stop, daily and total drawdown limit, kill switch, exposure cap.
Meta-labeling — a secondary model that lets through only signals with a real chance of profit.
Regime-aware weighting of 8 strategies based on market state (trend/range/volatility).
Hierarchical Risk Parity — capital split by correlations, diversified across bots.
Costs + slippage, Probabilistic & Deflated Sharpe — protects against a "lucky" backtest.
Automatic parameter optimization and rolling train/test windows against data-snooping.
US stocks + crypto under one roof. Crypto runs 24/7 including weekends.
Guides, glossary, how-to. Full audit log + CSV export for taxes.
New · headline feature
Caroline has a built-in conversational assistant of our own with access to your live account — portfolio, strategies, trades, signals, risk, engine status — plus the entire academy. Ask anything in natural language and get an answer grounded in your real data — and on your command it takes the action right away.
Answers and acts. It explains, guides, and on your command takes the action right away — creates a bot, runs a backtest, places a trade through your broker. Trades always go through your broker on your command, so you keep full control.
How it works
The engine wakes up, walks through your assets and repeats the whole chain. You don't have to be there.
Competitive moat
Four things no retail product offers at the same time today:
A PPO actor-critic that learns from a specific user's data. A new version goes live only if it passes an out-of-sample test; reward shaping accounts for transaction costs and loss aversion. Enterprise-grade ML lifecycle in a retail product.
Our own conversational assistant connected to the live account and the academy. You ask and you trade through chat — trading turns into a conversation. No other retail bot has this — and it's our onboarding and retention engine.
Meta-labeling lets through only signals with a real chance of profit; Hierarchical Risk Parity diversifies capital by correlations. Hedge-fund techniques, automated for retail.
No trade bypasses the risk manager: Half-Kelly sizing, kill switch at a 10% drawdown, trailing stop right at the broker, earnings blackout. Investors care about max loss — and here's the answer.
Market & opportunity
Alpaca Europe launched 2026 → a broker available to EU retail including CZ/SK.
AI/LLM mainstream → people now trust conversational assistants.
Retail boom → retail makes up ~25% of US stock volume (2× in a decade), $302bn of inflows in 2025 (+53% YoY).
EU Retail Investment Strategy → regulatory pressure to make investing more accessible.
The target reach is a fraction of the market → enormous room to grow. Institutional power in retail UX, no license and no coding.
Competitive analysis
Competitors only ever cover a slice. Caroline combines per-user learning, a conversational copilot, stocks+crypto and a risk engine under one roof.
| Platform | Stocks + crypto | Learning RL (per-user) | Conversational copilot | Risk engine (kill-switch/Kelly) | No-code + custom | EU-ready | Price/mo |
|---|---|---|---|---|---|---|---|
| Composer | ✓ | ✗ | ✗ | ~ | ✓ | ✗ | $ / free+ |
| Tickeron | ~ | ✗ | ✗ | ✗ | ~ | ✗ | $$$ |
| 3Commas | crypto | ✗ | ✗ | ~ | ✓ | ~ | $15+ |
| Trade Ideas | stocks | ✗ | ✗ | ✗ | ~ | ✗ | $$$ |
| eToro | ✓ | ✗ | ✗ | ✗ | copy | ✓ | % spread |
| QuantConnect | ✓ | ~ code | ✗ | ~ | code | ~ | $$ |
| Caroline | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | €0–99 |
Business model · B2C SaaS + credits
Subscriptions tiered on bots, symbols, the RL agent and the copilot quota. Risk management is full on every plan. Compute-heavy features (grid, copilot) on credits.
Optimization credits for grid search · copilot credits €5/500 messages · +3 bots (€9) · extra broker (€12) · deeper data history. Whoever actually consumes compute pays — it protects the margin.
B2C = an analytical and execution tool; trades go through the user's own broker on their command. Regulated services (portfolio management) via B2B partners (MiFID broker, CASP). The copilot is advisory in Phase 1 → safe.
Unit economics
Fixed cost is small: one shared AI model serves everyone; sentiment and indicators are computed once per symbol, not per user.
The only real variable cost is the copilot (LLM tokens). That's why every plan has a monthly message quota + credits. Same for grid search.
Result: ~80% margin that a customer can't unexpectedly overshoot.
24-month plan · conservative scenario
Cumulative loss peaks around −€13k at M7 (the deepest point). Then the business turns profitable and by M24 generates ~€600k+ of cumulative operating profit. At M24, net monthly profit is ~€60–65k.
We scale marketing + team from ~€5k/mo to ~€17k/mo — always below the growing revenue. Cloud grows in blocks of €80/100 users, marginally. Referral (15–20% of new users) pushes CAC down. Without B2B upside that would speed the curve up.
Investment & returns
Purely to operate, ~€50–60k would be enough. We take €300k strategically — faster growth, a team, a comfortable 24-mo. runway and room for the first B2B deals.
At M24: ARR ~€1M+, ~80% margin, a default-alive business. Entering at pre-seed (~€1.5–2M post-money) and SaaS ARR multiples of 5–10×, a 3–5× markup by Series A within ~24 months is realistic — for a business that is profitable by then.
This is not a guarantee of returns. The numbers are a conservative model without B2B upside.