ALEX: Separation of Decision Authority in an Autonomous Investment System — Deterministic Financial Intelligence, Governed AI, and Automated Trading

ALEX: Separation of Decision Authority in an Autonomous Investment System — Deterministic Financial Intelligence, Governed AI, and Automated Trading This technical report presents ALEX, an autonomous investment research and paper-trading platform designed around a fundamental architectural principle: artificial intelligence may assist financial reasoning and explanation without possessing direct authority over automated investment decisions or trade execution. Developed between April and October 2026, ALEX comprises approximately 211,000 lines of production Python code across 307 modules, 1,123 API routes, and approximately 4,000 automated test functions. Unlike architectures that rely on large language models (LLMs) to generate trading decisions, ALEX separates its quantitative investment engine, portfolio management, risk governance, and broker execution into distinct software components. The system incorporates a deterministic five-factor investment scoring model, rule-based portfolio construction, nine reference portfolios, and an execution architecture with 18 pre-trade risk controls. Automated orders are processed through Interactive Brokers in a strictly controlled paper-trading environment. A key architectural property is operational independence from generative AI: the automated fund's decision-to-order path does not require an LLM to select securities, determine portfolio weights, or submit trades. AI is primarily employed for research assistance, financial explanations, education, and conversational interaction. A separate AI-selected model-portfolio feature is explicitly documented as an exception and does not automatically control broker execution. The report includes practical examples of investment scoring, portfolio allocation, risk-based order rejection, financial-statement validation, and system behavior during LLM service interruptions. It also documents the platform's limitations, including incomplete decision reproducibility, data-quality dependencies, parallel quantitative implementations, ongoing security improvements, and the need for longer forward performance evaluation. The principal contribution is an implemented architecture for separating cognitive assistance, investment decision authority, and execution control within an autonomous financial system. ALEX does not claim to have invented algorithmic trading or demonstrated superior investment returns. Instead, it provides a practical engineering case study of how generative AI can be integrated into financial infrastructure without becoming the authoritative component responsible for automated trading decisions. Keywords: Autonomous Investment Systems; Financial Artificial Intelligence; Deterministic Investment Engines; AI Governance; Algorithmic Trading; Decision Authority; Quantitative Finance; Risk Management; Human-Centered AI; Financial Technology. Author: Teodor MinchevPublication type: Technical ReportVersion: 1.1Date: October 31, 2026License: Creative Commons Attribution 4.0 International (CC BY 4.0)