AI Quant Research and Risk Bootcamp: Building, Adapting and Governing Financial Language Models

Banks, asset managers and hedge funds increasingly want AI systems adapted to their own data, workflows and investment problems—not simply access to a general-purpose chatbot. This practical bootcamp examines how frontier-model APIs, hosted open-weight models and small language models can be combined with financial retrieval systems, structured market data and proprietary research. Participants learn how to design a grounded financial RAG pipeline, prepare data for supervised fine-tuning, understand LoRA and DPO, and evaluate whether model outputs contain useful and reproducible financial information. The workshop uses investment research, credit-risk and alpha-research examples. It addresses point-in-time data, vintage leakage, retrieval quality, prompt sensitivity, provenance, calibration and model drift. Semantic evidence graphs are used to trace how evidence and uncertainty move through the pipeline and to determine when an output is sufficiently supported for research or operational use. Participants leave with a practical architecture, evaluation framework and governance plan for a customized financial AI workflow. Schedule: • 9:30–10:15 — Financial language-model architectures: Compare frontier APIs, hosted open weights and SLMs. Decide when to use RAG, adaptation or both. • 10:15–11:15 — Building grounded financial RAG: Retrieve point-in-time news, filings and market data; inspect relevance, provenance and temporal validity. • 11:15–11:30 — Break • 11:30–12:30 — Post-training for financial applications: SFT, LoRA/QLoRA, preference data and DPO; what each method changes and when it is justified. • 12:30–1:15 — Lunch • 1:15–2:15 — Hands-on adaptation lab: Prepare a financial instruction dataset and compare a base model, RAG-augmented model and adapted model. • 2:15–3:00 — LLMs and SLMs for alpha and risk: Convert language outputs into testable signals; avoid look-ahead bias, overfitting and vintage leakage. • 3:00–3:15 — Break • 3:15–4:15 — Evaluation and evidence attribution: Test grounding, retrieval dependence, prompt robustness and residual model variation using semantic evidence graphs. • 4:15–5:00 — Deployment and governance clinic: Define release gates, human review, monitoring, retraining triggers and a plan for one participant-selected workflow. Your registration includes coffee on arrival, morning and afternoon refreshments, lunch, and a reception following the STAC Summit. During breaks and lunch, you’ll also have access to the STAC Summit exhibition hall—great opportunities to meet fellow attendees and connect with exhibitors. Location: New York Marriott Marquis, 5th Floor, 1535 Broadway. About the Instructor, Matthew Dixon, Ph.D., FRM: Matthew F. Dixon is a mathematician and quantitative finance researcher whose work lies at the intersection of machine learning, stochastic control, Bayesian inference, financial risk and the validation of agentic AI systems. He formerly served as Associate Professor of Applied Mathematics, Director of the M.S. in FinTech program and an affiliate professor in the Stuart School of Business at Illinois Institute of Technology. Earlier appointments include the University of San Francisco, UC Davis and Stanford University’s Institute for Computational and Mathematical Engineering. Before and alongside academia, he worked in quantitative roles at Lehman Brothers, the Bank for International Settlements, Barclays Capital and Silver Lake Partners. He holds a Ph.D. in Applied Mathematics from Imperial College London and was named Risk Magazine’s Buy-Side Quant of the Year in 2022.