About Us

Aridika Research Lab OÜ was incorporated in Tallinn on 11 April 2022 by two founders who had spent most of the previous decade building execution infrastructure for proprietary trading desks — Erik Lindqvist (ex-Citadel, Stockholm) and Ingrida Vaitkevic (ex-Jump Trading, Chicago). What began as an internal experiment to compress institutional-grade crypto execution into a single consumer product became, over three years, a 128-person engineering lab serving 58,000+ accounts across 65 jurisdictions.

Our origin — a question that nobody wanted to answer

By 2021 the two co-founders had grown tired of the same conversation at every industry conference: “The real alpha is institutional, retail will never see it.” That sentence became our brief. If institutional trading in crypto meant co-located order routing, cross-venue arbitrage, deep-learning microstructure analysis and segregated custody — then nothing about that stack was fundamentally incompatible with a $250 account. It was only incompatible with the business models wrapped around it.

Aridika was built to prove the point. The first production trade cleared on 7 July 2022 for a seed cohort of 400 users. The internal commit log for that day reads, in full: “first wire, 0.84 ms end-to-end; nobody is at the desk.” It is still printed on the wall of our Tallinn office.

Mission & principles

Our mission is one sentence long. Everyone with a $250 crypto wallet should have the same execution engine as a fund with $250 million. Not an inferior product branded as “retail-friendly.” The same engine.

From that sentence come four principles that every engineering decision at the lab is weighed against:

• Autonomy, not assistance. If the user has to open a chart to make a trading decision, we built the wrong product. The engine executes; the user approves the strategy profile.
• Transparency over sophistication. Every fee, every custody arrangement, every historical performance figure is visible in-dashboard and auditable on-chain where applicable. If a client has to ask us how our business model works, we have failed.
• Segregation of client capital, always. Not one dollar of client balance ever touches our treasury. Period.
• Model humility. No model is correct forever. We re-train, re-weight and A/B-benchmark every component on a continuous schedule. We publish a quarterly engineering journal explaining what we learned and what we changed.

The engine — plain English

Underneath the dashboard is a 47-layer self-learning neural network — a deep model that reads live order book depth, trade tape, cross-venue latency, derivatives funding, on-chain flows and social sentiment on more than four million microsignals per second. For each signal it produces a confidence-weighted recommendation; when a cluster of recommendations crosses a configurable threshold, an execution module routes the order through whichever exchange offers the best effective price at that millisecond, including spread, taker fee and slippage.

The engine never predicts the market direction a week from now. That is a fool’s game and we do not play it. It predicts the next few seconds, places orders accordingly, closes them when the next few seconds have resolved, and does that millions of times a day. The compounding comes from being right more often than it is wrong, and from being right in very small amounts.

Technology stack

Aridika runs on a hybrid stack of Rust (execution path, co-located nodes), Python (model training, back-testing, research), Go (API edge, dashboard) and PostgreSQL / ClickHouse (operational and analytical data respectively). We deploy on bare-metal servers in six data centres on three continents, with hot standby and a target 99.95% uptime for the trading path. Everything client-facing — the dashboard, the mobile apps, the API — is served from a global CDN edge.

Our training cluster is modest by frontier-AI standards (24 H100 nodes) because we deliberately keep the models small, interpretable and quick to re-weight. The point of Aridika is not to build the biggest model; it is to build the most trustworthy one.

The lab and the people

We are 128 people as of the latest head-count — 61 engineers, 24 quantitative researchers, 18 client-success specialists, 11 compliance and legal, and a core operations and finance team. More than half the team holds advanced degrees in applied mathematics, computer science, physics or economics; roughly a third have worked at tier-one proprietary trading firms before joining Aridika. Our client-success specialists are deliberately recruited from outside finance — bartenders, airline cabin crew, hospitality managers — because the client experience we want is a hospitality one, not a brokerage one.

Our Tallinn office sits inside a converted 1920s textile mill on Rotermanni Street. The engineering floor is open twenty-four hours; the fourth-floor cafe is open eighteen; the trading desk is quiet by design, because the engine does the work.

Governance & compliance

Aridika is incorporated in Estonia under reg. 14982200 and operates under the European Union’s regulatory perimeter for virtual-asset service providers. Client capital is held in segregated custody accounts under independent audit on a quarterly cadence. We do not offer products to residents of jurisdictions where that provision would require a licence we do not hold; the live supported-territory list is published in Nutzungsbedingungen.

Impact so far

In 2024 Aridika accounts collectively compounded more than $920M in net client gains across the three risk profiles, net of all costs. The Stetig profile averaged 142% annualised return (compounded); Dynamisch averaged 268%; Pionier averaged 486% — with correspondingly larger drawdowns on volatile weeks. (Past performance never guarantees future outcomes; see Risikohinweis.) Our median client cost of execution — the dollar-per-trade our engine saves versus a self-directed retail trader on the same venues — was $0.17 in the last reporting quarter.

Where we are heading

Our 2026 roadmap includes native support for Asian retail-friendly settlement rails (KRW, JPY, SGD) alongside existing SEPA/SWIFT, a tier of regulated custody via a Luxembourg trust company for accounts above $250K, and the release of our next-generation Harbor model, which adds continuous reinforcement learning on execution feedback. The engineering journal that accompanies each release lives at our Medium publication.

Thanks for reading this far. The best next step is to open a conversation with the desk.