STAC New York - LLM/Generative AI for Systematic and Discretionary Trading - NVIDIA

Sep 4, 2026

STAC New York presentation outlining how LLMs and generative AI can be applied to systematic and discretionary trading using alternative and unstructured data. It frames market volatility and difficult behavioral-finance signals as core problems, then maps solutions across data ingestion, NLP-driven alpha and beta extraction, and LLM-generated trading/investment signals. The content also details an end-to-end workflow for knowledge metrics aggregation, downstream indices, and evaluates AI ROI via workshops, NeMo Foundry prototyping, testing, and deployment.

LanguageEnglish

CategoryTechnology

FormatPDF

Pages6

Reading Time15 min

AccessFree to download

Application frameworks platform

NVIDIA HPC

NVIDIA AI Enterprise

NVIDIA Omniverse

System software and accelerated hardware

Why LLMs/Generative AI in capital markets

Staying on top of market: problems and consequences

Industry needs and solutions

Workflow for LLM-driven trading

Step 1: Use unstructured alternative data

Step 2: Generate trading signals with LLMs

Step 3: Aggregate knowledge metrics by ticker and themes

Step 4: Downstream investing and trading uses

Getting started and measuring AI ROI

Workshops and problem definition

Prototype with NVIDIA LLM NeMo Foundry

Test, document, and present ROI

STAC NEW YORK LLM /GENERATIVE AI FOR SYSTEMATIC ALGO/DISCRETIONARY TRADING INCL . ALTERNATIVE DATA BY PRABHU RAMAMOORTHY, NVIDIA FINANCIAL SERVICES & TECHNOLOGY TEAM CFA, FRM, CAIA NVIDIA AI LEADER FOR LLM/GENERATIVE AI 2 APPLICATION FRAMEWORKS PLATFORM NVIDIA HPC NVIDIA AI ENTERPRISE NVIDIA OMNIVERSE SYSTEM SOFTWARE ACCELERATED HARDWARE RTXUCF RTX GPU DOCA DGX CPU MAG HGX DPU CUDA-X BASE CMD EGX OVX NIC FLEET CMD SUPER POD SWITCH PHYSX AERIAL AGX SOC Full Stack-Software + Hardware Data Center Scale (Cloud/Onprem) 2,700 Accelerated Applications 450 SDKs, AI Models WHY LLMs/GENERATIVE AI in Capital Markets Used in Systematic Trading/Discretionary Trading McKinsey reports -Advanced analytics in asset management beyond the buzz, Generative AI is here to change biz |
Problem statementStaying on Top of Market
Results and Consequences
Industry Needs & Solutions | | | | | | | --- | --- | --- | --- | --- | --- | | ▪ Extremely Volatile Market | | ▪Lower Sharpe -> Redemptions | | Needs | | | ▪Market signals are viral – | | ▪Downside Risk - Failures such | | ▪Continuous Real time exposure | | | Behavioral finance is hard | | as Credit Suisse, SVB, Banks, | | ▪Unstructured -> Alpha/Beta | | | ▪Unstructured potential is not | | Melvin Capital HF (Game stop) | | Solutions | | | tapped fully | | ▪ Upside - Unable to generate excess returns/beat market | | ▪NLP Older -> Newer accuracy
▪ Early Warning Indicators | | |
Step 1 – Use Unstructured Data – Alternative data
Alpha, Beta
Step 2-Generate Investment/Trading Signals with LLMs
Step 3- Knowledge Metrics Aggregated by Ticker & Themes
Step 4-Downstream uses – Investing/Trading Indices, Systematic/Discretionary
3
| | | | | | Leverage Accelerated workloads in Capital Markets for you One Stop Platform -Not only LLMs but use it with Quant Finance, ETL/ML, DL Algos AI (Neural nets) – LLMs/Generative AI/others Quant Finance Aka HPC Data Processing ETL/ML (Non neural nets) • AI Unstructured Data using NLP with LLMs, Other Systematic Trading Algos • Framework –PyTorch/TensorFlow, NVIDIA NeMo LLM, NVIDIA RIVA • Pricing, Risk (MC Sim, Margin, FRTB, CVA, SIMM, XVA) & Back testing • Framework – CUDA C/C++, Parallel Algorithms C++, NVIDIA Accelerated Python – RAPIDS, Open ACC • Feature Engineering, Data Prep, & Data Science (e.g. , XGBOOST) • Framework – NVIDIA Accelerated Python - RAPIDS, Spark on GPU Forrester Wave ™ : AI Infrastructure, Q4 2021- “NVIDIA’s DNA is in every other AI infrastructure solution we evaluated. It’s an understatement to say that NVIDIA GPUs are synonymous with AI infrastructure.” 4 How to Get Started With LLMs/Generative AI at your firm Generate your AI ROI McKinsey Modelling Impact of AI-Front-runner companies could “Double their cash flow” or NVIDIA LLM Nemo Foundry On DGX Cloud NVIDIA LLM Nemo tuning, guard – Prompt rails Test/Deploy on Platform of choice NLP- News, Docs Sentiment & Q/A Alt Data LLM/Generative-> AI ROI • Step 1: Workshops to prioritize,& define the problem statement, success criteria in LLM/Generative AI (2-4 hours) • Step 2: Leverage NVIDIA LLM NeMo Foundry on cloud to prototype with your proprietary data (2-3 weeks) • Step 3: Test & document (1 weeks) • Step 4: Present findings & ROI to key stakeholders (1 week) What Would an LLM/Gen AI Journey Look Like? THANK YOU

FAQs

How does the document position LLMs and generative AI for capital markets trading?

It explains that LLMs and generative AI can convert unstructured alternative data into actionable investment and trading signals, supporting both systematic and discretionary trading use cases.

What problems in the market does the framework address?

It highlights extreme market volatility, viral market signals, difficulty in behavioral-finance signal extraction, and challenges in downside risk and generating excess returns.

What are the recommended steps to get started at a firm?

The document proposes workshops to define the problem and success criteria, prototyping with NVIDIA LLM NeMo Foundry on cloud using proprietary data, then testing/documenting and presenting findings and ROI to stakeholders.

Sep 4, 2026

Upload by นรินทร์