AWS re:Invent 2025 - Trading innovation: Jefferies' AI assistant on Amazon Bedrock (IND3315)

AWS re:Invent 2025 - Trading Innovation: Jefferies' AI Assistant on Amazon Bedrock

Overview

Jefferies, a 60-year-old full-service investment bank, has partnered with AWS to build a trade assistant agent that leverages large language models (LLMs) and cloud-native technologies to empower its equity traders. This solution aims to address the challenges faced by traders in accessing and analyzing the vast amounts of real-time trading data to generate valuable insights.

The Challenge

• Traders at Jefferies struggle to access and analyze the massive amounts of trading data due to its fractured nature, stored across multiple data stores and visualization tools.
• Traders lack the time and technical expertise to build and maintain systems capable of delivering the insights they need to make informed trading decisions.
• The result is that traders face barriers that slow down their decision-making and impact both trading and client advisory capabilities.

The Solution: Jefferies' Trade Assistant Agent

• The trade assistant agent is built on AWS services and leverages LLMs, specifically the Titan embeddings model, to provide traders with a conversational interface to access real-time trading insights.
• When a trader submits a query, the underlying LLM generates the appropriate SQL query, which is then executed against the relevant data sources, such as Jefferies' in-memory data grid, GridGain.
• The response is then processed and presented to the trader in a visually appealing format, including charts, tables, and textual insights.
• The solution maintains conversational context to provide relevant suggestions and enable deeper analysis throughout the user's session.

Key Features and Architecture

1. Conversational Analytics Interface: Traders can interact with the assistant using natural language, and the system maintains context to provide relevant insights and suggestions.
2. Strands Agents: The solution uses Strands agents, which enable Jefferies' technical teams to build and run AI agents with minimal code.
3. Flexible LLM Integration: The architecture allows Jefferies to easily choose different LLMs for specific use cases through Amazon Bedrock.
4. Security and Compliance: The solution includes advanced guardrails and row-level data entitlements to prevent accidental access to sensitive customer data, with complete audit trails to meet compliance requirements.

Impact and Results

• Jefferies' beta rollout to 50 users across sales and trading operations resulted in an 80% reduction in the time spent on routine analytical tasks, unlocking significant efficiency and revenue generation capacity.
• The solution has reduced the technical burden for producing custom dashboards across multiple trading desks, with self-service capabilities that decrease the dependency on technical resources.
• Democratization of data access has enabled business users to query millions of records of equity trading data using natural language, allowing for real-time discovery of trading patterns and market opportunities.
• The architecture is designed to be future-proof, with self-learning capabilities that continuously improve and adapt, and seamless integration with Jefferies' existing BI platforms and infrastructure.

Key Learnings and Next Steps

1. Avoid LLM-generated visualizations: Jefferies found that using a Python library and a faster data store like an in-memory database is more effective than relying on LLMs to generate visualizations, due to the risk of hallucinations.
2. Multi-product expansion: Jefferies plans to extend the trade assistant beyond equities to support other product types and trading desks.
3. Global deployment: Jefferies aims to bring the proven efficiency gains of the trade assistant to its international trading operations.
4. Enhanced governance: Jefferies will strengthen observability and audit capabilities to meet regulatory and compliance requirements.
5. Advanced code generation: Jefferies is exploring the transition from a UI-based approach to sophisticated NLP-driven code generation for a better user experience.
6. Reusability and firmwide adoption: Jefferies is identifying similar opportunities across other business areas and aims to turn the trade assistant solution into a generic API that can be used across the firm.

Conclusion

Jefferies' partnership with AWS to build the trade assistant agent on Amazon Bedrock has resulted in a powerful solution that addresses the challenges faced by equity traders in accessing and analyzing real-time trading data. The solution's conversational analytics interface, flexible LLM integration, and focus on security and compliance have delivered significant efficiency gains and unlocked new opportunities for Jefferies' trading operations. The key learnings and future plans outlined in this presentation demonstrate Jefferies' commitment to continuous innovation and its ability to leverage cloud-native technologies to empower its traders and better serve its clients.