Perceive
Establish controlled data, point-in-time features, provenance, and reproducible evidence.
Four essential texts · Thirteen notebooks · One governed architecture
Algorithmic Trading in the Age of Artificial Intelligence
Build the institution. Govern the intelligence.
From quantitative trading strategies to autonomous AI research systems. Explore the book, executable notebooks, QuantConnect implementation, and governed assistant that connect market data, machine learning, portfolios, execution, and independent risk.
The architectural premise
Intelligence needs
an institution.
A prediction is only the beginning. A useful autonomous system must know which evidence it can trust, which procedures it may invoke, how to challenge a result, when to stop, and who remains responsible. This project makes those obligations part of the design.
Reason
Compare models and strategies. Use bounded LLM planning to select approved research capabilities.
Execute
Translate proposals into portfolios and backtests with explicit costs, timing, and constraints.
Govern
Preserve independent risk, falsification, audit, authority boundaries, and human judgment.
The journey · Five transformations
From 400 tasks
to an invokable
research institution.
The project began as a formidable engineering inventory. Each transformation made it easier to understand, reproduce, and use—while preserving the discipline underneath.
01
Define
Map the complete system
More than 400 tasks exposed the breadth of the challenge.
What this made possible
A comprehensive design spanning quantitative methods, operations, controls, and accountability.
02
Organize
Build in ten days
An execution program made the engineering effort manageable.
What this made possible
A documented diary, staged deliverables, and an installation-oriented implementation.
03
Teach
Make the method executable
The work became a cumulative notebook curriculum.
What this made possible
Inspectable experiments, reusable capabilities, and an integrated autonomous research process.
04
Translate
Cross the platform boundary
NB07 carries the research design into QuantConnect/LEAN.
What this made possible
A governed specification becomes an event-driven research backtest project.
05
Invoke
Encode the research discipline
The notebook methodology becomes an assistant operating specification.
What this made possible
Canonical instructions, states, evidence gates, installation files, and a runtime scaffold.
The reading collection
Four texts.
One cumulative argument.
Enter through the conceptual overview, follow the executable methodology, then explore the two implementation bridges. Each text addresses a different question in the same architecture.
Think. Build.
Use and improve.
The complete conceptual overview of an autonomous algorithmic trading system, connecting traditional quantitative finance with tools, skills, agents, governance, and the ten-day engineering record.
Read the book ↗Building an autonomous quantitative research system
A pedagogical explanation of NB00–NB10: why each notebook exists, how its methods work, and what it contributes to the institution.
Read the notebook paper ↗From research
to QuantConnect / LEAN
A thorough analysis of NB07 and the translation from Colab research to event-driven execution, including model competition, portfolio construction, risk, and reconciliation.
Read the QuantConnect paper ↗From notebooks
to a governed assistant
The final transition: preserve the research discipline in an invokable assistant, with a canonical specification, a worked mission example, and detailed installation guidance.
Read the assistant paper ↗The executable collection · 13 notebooks
Learn the system
by building it.
Begin with NB00A, NB00, and NB00B. Continue through NB01–NB06, then NB08–NB10. Study NB07 separately after the integrated research sequence, so platform engineering does not interrupt the learning progression.
01 · Foundations
Establish the world and the first complete decision
Autonomous Systems Protocol
Define the constitution, schemas, permissions, evidence, and authority boundaries.
Synthetic Equity Market Engine
Create a reproducible market with regimes, sectors, corporate actions, and known data defects.
Minimal KNN Vertical Slice
Follow one complete chain from governed data to predictions, positions, costs, stress, and audit.
02 · Capabilities
From quantitative methods to a coordinated institution
Models, Strategies, and Portfolios
Compare heterogeneous models, trading rules, and portfolio methods under common evaluation discipline.
Execution, Risk, and Audit
Introduce simulated orders and fills, reconciliation, implementation costs, stress, and independent risk.
Tools, Skills, and Bounded Agents
Package analytical functions as discoverable capabilities with explicit contracts and permissions.
Governed Agent Constellations
Coordinate research, strategy, portfolio, and risk roles while preserving responsibility and dissent.
The Governed Meta-Agent
Decompose missions, discover approved capabilities, route work, and manage bounded resources.
The Closed-Loop System
Integrate admission, dynamic planning, execution, monitoring, recovery, and audit closure.
03 · Integration
Bring quantitative tools and LLM reasoning together
Integrated Quantitative Research
Connect actual model execution, strategy comparisons, portfolios, and risk to dynamic orchestration.
The Bounded LLM Control Plane
Add real language-model planning and evidence critique around deterministic quantitative execution.
The Capstone Institution
Rejoin the canonical data contract with integrated research, LLM reasoning, stress, and adversarial tests.
04 · Platform bridge
Preserve the method inside QuantConnect
The QuantConnect / LEAN Bridge
Translate the research specification into an event-driven backtest with costs, independent risk, and provenance.
The discipline that survives
Autonomy within
an explicit
mandate.
The LLM proposes and interprets. Quantitative tools calculate. Independent risk challenges. Governance constrains. Humans decide.
Evidence before claimsKeep data, assumptions, configurations, and results traceable. Missing evidence remains missing.
Authority before actionA planner may choose approved tools; it cannot grant itself new permissions.
Challenge before promotionStress, falsification, and independent risk determine what survives the research process.
Accountability throughoutResearch completion, platform validation, and deployment authorization remain separate decisions.
Two implementation routes
Take the method
into practice.
Use the platform package to study the generated backtest. Use the assistant bundle to install the research workflow or develop its executable runtime. The available evidence and remaining integration work are explicit.
From specification
to platform behavior.
A reference implementation with 30 U.S. equities, five competing model families, monthly training, cross-sectional long/short signals, portfolio constraints, and audit artifacts.
Start with the package guide, add the supplied algorithm and configuration to a QuantConnect Python project, and run a backtest. Reconcile platform behavior with the research contract.
Current evidence
Local checks are recorded as passed. QuantConnect compilation and backtesting are recorded as NOT_RUN. Live mode is denied.
Explore the QuantConnect package ↗Local checks are recorded as passed. QuantConnect compilation and backtesting are recorded as NOT_RUN. Live mode is denied.
State the mission.
Preserve the discipline.
An autonomous research assistant configured within a dedicated ChatGPT Project. Canonical instructions and knowledge files establish its governed workflow. States S0–S11 connect mission admission, research evidence, independent risk, critique, LEAN packaging, and audit closure.
Follow the installation manifest and acceptance tests. Developers can extend the accompanying Python runtime with validated notebook capabilities.
Current implementation
The ChatGPT bundle provides procedural guidance. The Python runtime is a scaffold: quantitative handlers and an operational LLM adapter still require integration.
Explore the assistant bundle ↗The ChatGPT bundle provides procedural guidance. The Python runtime is a scaffold: quantitative handlers and an operational LLM adapter still require integration.
An illustrative first mission
“AIAT mission: compare a 20-day momentum rule with approved predictive models in the governed synthetic-equity universe. Preserve chronological validation, transaction costs, independent risk, and audit evidence. First identify the current state, missing artifacts, and the next permissible transition. Live authority: none.”
Choose your starting point
One collection.
Several ways in.
The materials support conceptual study, hands-on experimentation, platform implementation, and assistant development. Start with the question you want to answer.
Understand
Study the complete conceptual architecture and the reasoning behind its controls.
Experiment
Execute the notebooks and inspect the assumptions and evidence at each stage.
Implement
Understand how a governed research design maps to event-driven platform behavior.
Invoke
Install the conversational workflow and test whether it preserves the methodology.