Menu
Portfolio · AI decisions in operation

AI Trading.
Five systems. Distinct AI roles.

AI traders, supervised bots, plug-in strategy sleeves, broker execution and stock screening with performance tracking.

Explore each system’s model role, architecture, supporting platform and process. Trading examples demonstrate engineering; they are not investment advice or a performance guarantee.

Trading terminalMarket view
Market-chart excerpt from the OANDA terminal, without account, position or order information
Spartan overviewDemo data
Spartan overview interface demonstration showing system health, sleeve signal parity and illustrative positive figures; not live trading results
View full screen
Spartan PulseDemo data
Spartan Pulse interface demonstration with illustrative balance of $125,615.86 and day P&L of +$1,125.00; not live trading results
View full screen
AI judgement Risk controls Broker execution
Spartan interface demonstrations · illustrative balances and P&L, not live results
Portfolio family

Explore the architecture. Follow the process.

This is not one generic trading bot. Each implementation gives AI a different job, while deterministic services retain control over risk, broker access, execution truth, recovery and evidence.

System architecture + process

Apex Trader · the AI model acts as trader.

The model combines market, news, macro and portfolio evidence, chooses the action and explains its rationale before controlled tools execute it.

Operator experienceDashboard + Telegram · operator visibility, confirmation and alerts
AI / model roleApex AI trader · contextual judgement · memory and case context
Decision systemEvidence synthesis · enter, wait, manage or exit · method and position management
Authority + riskTiered authority: automatic, confirm-first or alert-only · risk and conflict gates
Execution + brokersMoomoo · IBKR · Tiger · Alpaca
Evidence + learningOrder lifecycle · fills · protection · SQLite audit · watchdogs and recovery
Technical platformPython services · FastAPI · Telegram · SQLite ledgers · broker adapters
How it operates

AI interprets. Deterministic services protect the action.

Evidence is preparedBars, positions, news, macro and event context are checked for freshness and conflict.
AI trader decidesThe model chooses enter, wait, manage or exit and explains why.
Authority and risk gateThe configured tier determines whether the action is automatic, confirmed or alert-only.
Broker truth returnsThe authorised venue executes; fill, protection and reconciliation become the official record.
MoomooIBKRTigerAlpaca

AI owns the trading judgement; calculations, authority, execution, fill confirmation, protection and audit remain deterministic.

System architecture + process

Vega Trader · AI judgement over a fast deterministic controller.

Vega chooses a trading method and intent; the controller handles market feeds, position sizing, execution, circuit breakers and rapid protection.

Operator experienceVega console + dashboard · authority, visibility and intervention
AI / model roleVega AI trader · retained lessons, case continuity and anti-drift context
Decision systemMethod selection · trend, breakout, range, bounce and recovery methods
Authority + riskIntegrity and drift gates · explicit live authority · fast circuit breakers
Execution + brokersOANDA
Evidence + learningLedger · broker mirror · reconciliation · retrospection and approved lessons
Technical platformPython controller and AI agent · OANDA v20 · persistent context and ledgers
How it operates

The AI pilots. The fast controller protects.

ObserveOANDA feed and current broker state become a decision packet.
Vega decidesThe AI selects a method and issues entry, hold, manage or exit intent.
Controller actsSizing, order placement, brackets and tick-level protection run deterministically.
DebriefThe outcome is reconciled and approved lessons are retained.
OANDA

The model is the trader, while the controller reacts faster than model latency and preserves broker truth.

System architecture + process

Spartan Trader · plug-in strategy sleeves with AI research and AI trading.

Independent, evidence-backed sleeves can be added, disabled or replaced without rewriting siblings. AI helps research sleeves and acts as the head trader over the approved sleeve book.

Operator experienceOwner console · portfolio evidence · release and authority decisions
AI / model roleAI research partner · AI head trader over the approved sleeve portfolio
Decision systemIndependent sleeve plug-ins · stable identity · portfolio compiler · lifecycle registry
Authority + riskReplay and hostile audit · immutable promotion · risk and admission rules
Execution + brokersOANDA
Evidence + learningExact bar journal · per-sleeve receipts · reconciliation · immutable release evidence
Technical platformPython engine · SQLite operational truth · React / Vite console · OANDA APIs
How it operates

Build specialist sleeves. Let AI lead the approved team.

Sleeve factoryAI research → independent strategy module → replay and hostile audit → owner promotion.
Live sleeve bookSpecialists evaluate → AI head trader interprets → risk admits → OANDA executes and reconciles.
OANDA

Sleeves are loosely coupled and versioned; identity, portfolio composition, risk, fills, recovery and release remain controlled.

Explore the Spartan consolesOverview and Pulse · illustrative demo figures
Spartan overviewDemo data
Spartan overview interface demonstration showing system health, sleeve signal parity and illustrative positive figures; not live trading results
View full screen
Spartan PulseDemo data
Spartan Pulse interface demonstration with illustrative balance of $125,615.86 and day P&L of +$1,125.00; not live trading results
View full screen

Interface demonstrations only. Balances and P&L are illustrative, not actual account figures or evidence of trading performance.

System architecture + process

Multibroker · two distinctly different AI roles.

One operating platform runs either with Claude as the trader, or with a deterministic bot trading while Claude supervises health, drift and exceptions.

Operator experienceFleet dashboard · alerts · account and engine visibility
AI / model roleClaude trader · Claude supervisory manager
Decision systemMode A: contextual AI decision · Mode B: approved deterministic strategy
Authority + riskPer-instance permission · fail-closed checks · deduplication · isolation
Execution + brokersTiger · IBKR · Alpaca · Moomoo
Evidence + learningOrders · fills · account truth · health · exception history
Technical platformPython · Streamlit / Plotly dashboard · broker SDKs · shared market-data hub
How it operates

Two desks. One guarded infrastructure.

Claude tradesMarket and account evidence → Claude judgement → risk gate → authorised broker execution.
Bot trades; Claude supervisesApproved strategy engine → controls and orders → Claude watches drift, health and exceptions → people can halt or recover.
TigerIBKRAlpacaMoomoo

Both modes share broker adapters, account isolation, risk gates, durable ledgers, fill confirmation, restart recovery and reconciliation.

System architecture + process

AI Stock Screening · AI research with measured outcomes, not order authority.

Deterministic screening qualifies candidates; Qwen explains the case, Kronos contributes a shadow forecast, and later observed outcomes test whether the methods were useful.

Operator experienceScreening dashboard · candidate lifecycle · evidence · watchlists · performance history
AI / model roleLocal Qwen analyst · Kronos shadow forecasting · bounded web-assisted context
Decision systemDiscovery and ranking · deterministic eligibility · AI may veto but cannot promote a failed candidate
Authority + riskLiquidity · freshness · activity · evidence gates · controlled watchlist permissions
Execution + brokersMoomoo
Evidence + learningSaved scenarios compared with later observed bars · hypothetical outcomes, not broker fills
Technical platformFastAPI · React / Vite · SQLite history · local Qwen · Kronos · Moomoo OpenD
How it operates

Screen. Explain. Measure what happened later.

Discover and qualifyMarket activity is screened; weak or incomplete evidence is removed.
Qwen and Kronos researchQwen builds the thesis; Kronos adds an experimental shadow forecast.
Human reviewThe case is presented as decision support without implying certainty.
Track outcomesThe scorekeeper compares the saved scenario with later observed market behaviour.
Moomoo

Models have no broker order tools. Performance tracking evaluates research methods using observed hypothetical outcomes.

Architecture

Loosely coupled components

Models, strategy sleeves, brokers and controls can evolve independently.

Operations

Consequential automation

Execution requires observable risk, reconciliation, monitoring and recovery.

Client relevance

AI inside a controlled system

The same approach applies wherever AI influences material operational decisions.

Contact us

Which operating problem should we start with?

Tell us what is slowing the work down, what you want to change and which people or systems are involved. We can discuss a Blueprint, a pilot, an implementation or a defined enterprise technology mandate.

The problemWhat is happening today?The outcomeWhat should improve?The environmentWho and what is involved?
Singapore
Malaysia