ai ml
Explore core specifications, mathematical structures, and research insights categorized under ai ml.
Agent as a Compiler Framework
Architectural evaluation of the 'Agent as a Compiler' framework for quantitative finance: replacing sequential ReAct loops with decoupled Chat and Coding agents, typed Analysis Plan IRs, O(1) DAG parallelization, AST-enforced semantic equivalence, and an automated continuous learning 'Teach' flywheel.
Vector Storage Solutions for Confluence RAG
A feature-by-feature comparison of Chroma, FAISS, and Scikit-learn for hierarchical Confluence RAG — Chroma wins on architectural fit and end-to-end latency for filter-heavy queries despite FAISS's raw speed advantage.
Database Agents with MCP and LangChain
Architecting production-grade database agents by fusing MCP (standardized tool communication) with LangGraph (stateful orchestration), covering context provisioning strategies and a defense-in-depth security posture across database, application, and LLM layers.
The Dual-Purpose Playbook: Confluence for Human and AI
How to architect a Confluence knowledge base that's equally usable by humans and AI/RAG systems, via five principles: Architect, Atomize, Structure, Automate, Govern — plus a no-code database pattern using Page Properties macros.
Ollama Cheat Sheet: Complete Command Reference
A quick-reference for running LLMs locally with Ollama — model management, interactive chat, Modelfiles, the local HTTP API, and advanced tips like tool calling and Open WebUI.
Architecting Advanced RAG Systems: Metadata-Driven Filtering
Fusing semantic vector search with structured metadata filtering — pre- vs. post-filtering tradeoffs, vector database comparisons (Qdrant, Pinecone, Weaviate, pgvector), self-querying retrieval, RAG vs. NL-to-SQL, and secure multi-tenant RAG design.
Architectures of Intelligence: Advanced RAG and Context Engineering
A tiered framework for production RAG systems — chunking strategies, query transformation, two-stage re-ranking, prompting patterns, and agentic self-correction loops (CRAG, SELF-RAG) — organized around 'context failures, not model failures.'
LSTM in Systematic Trading: Architecture, Application, and Performance
How LSTM networks solve the vanishing gradient problem to capture long-term dependencies in noisy, non-stationary financial time series, compared against GRU and Transformer architectures and the practical pitfalls of deploying them.
Transformers in Systematic Trading
How Transformers adapt to finance via time-series patching, applications in forecasting/NLP/factor generation, a head-to-head comparison vs. LSTM and XGBoost, and case studies (Stockformer, Quantformer).
XGBoost vs. Deep Learning in Systematic Trading
Why XGBoost remains the right tool for structured, tabular, feature-driven prediction problems (cross-sectional ranking, regime classification) even as deep learning advances, plus the hybrid LSTM-then-XGBoost architecture pattern.
Why Social Media Recommender Algorithms Can't Pick Stocks
Why engagement-optimized recommender systems (TikTok-style) are structurally incompatible with sound financial advice, the regulatory risks of applying them to markets, and the viable path forward (educational augmentation, not prescriptive recommendations).
Reinforcement Learning in Quantitative Trading
How RL shifts trading from predict-then-act to directly learning a cost-and-risk-aware policy, its core application domains (portfolio optimization, execution, market making), practical limitations, and a blueprint for building an RL trading system.
Why Machine Learning Assumptions Break in Financial Markets
How non-stationarity, volatility clustering, and fat tails violate the core assumptions behind ML algorithms in financial markets, and how tree-based vs. deep learning models each handle (or fail to handle) these violations.
The Evolution of Deep Learning in Quantitative Trading
A comprehensive technical survey charting the evolution from traditional econometric models to sophisticated deep neural networks in quantitative finance. Explores MLPs, LSTMs, CNNs, Autoencoders, Deep Reinforcement Learning, GNNs, and Transformers—analyzing their unique properties, applications in trading, and critical limitations in high-noise, non-stationary financial markets.
Agentic RAG with LangChain
A technical guide to building agentic RAG systems that integrate LangChain with custom proprietary wikis — multi-hop reasoning, math-aware chunking, LangGraph workflows, and hierarchical agent swarms for quantitative finance.
The Evolution of Autonomous Execution
A comprehensive technical deep-dive into the evolution of AI tool-calling architectures in quantitative finance.
Claude Code Financial Cheatsheet
A comprehensive enterprise reference for using Claude Code in quantitative finance workflows.
Building Interactive Financial Copilots
A comprehensive architectural masterclass on designing Generative UIs for financial dashboards.
Formulaic Alpha Mining & Deep Search
A comprehensive guide to automated alpha discovery using Deep Reinforcement Learning and Monte Carlo Tree Search. Master formulaic operators, risk-seeking policy gradients, and the Deflated Sharpe Ratio to separate true structural alpha from backtest overfitting.