In Part One of this series, we introduced the Alpha Assistant, an agentic AI system designed to accelerate quantitative research by bridging the gap between generic large language model (LLM) responses and a fully integrated output, actionable within the firm’s proprietary research environment. While powerful, we highlighted that the Alpha Assistant faces limitations inherent to current LLM architectures: LLMs can only pay attention to so much information in their context and can only work on tasks for so long before exhausting their available attention span. This left us with a key question: is there a better approach to overcome current model limitations?

In Part Two, we argue that the answer lies in specialisation. Rather than building a single, maximally flexible assistant, we can construct purpose-built agentic workflows optimised for particular research tasks. Enter AlphaTrend: a specialised agentic system designed for a specific area of trend-following signal research. Where the Alpha Assistant serves as an interactive coding partner, AlphaTrend functions as a predefined and structured autonomous research pipeline, systematically generating, implementing and researching signal proposals through a pre-defined workflow.

In what follows, we explore AlphaTrend’s architecture, demonstrate its capabilities with a practical example, and share what we’ve learned from deploying specialised agentic workflows in quantitative research.

Read | A Trend Following Deep Dive

Man Group

London-based Man Group is a technology-empowered active investor across equity, credit, commodities and currency markets, with a firm focus on delivering performance and client portfolio solutions.

Information provided is general information and is for educational purposes only, and does not include, or constitute as, financial product advice.