Title: AI-based intelligent assistant for developing and researching quantitative investment strategies with applicability to diverse business areas.
Acronym: LLM-GenFramework
This project aims to create a unique solution based on Large Language Models and Retrieval Augmented Generation methods to support the creation, research and analysis of quantitative investment strategies, and also applicable to other areas. Based on the successes of LLM models and RAG methods, the solution will increase labour productivity in the creation of quantitative investment strategies, and in other areas related to data analysis and IT solution development.
We will create a generic LLM-GenFramework module to apply the above-mentioned techniques to various business areas, as well as a specialised LLM-Invest product designed to support the research, development and analysis of investment strategies.
This Project will research and develop the architecture of a complete, integrated system using RAG techniques for use as a knowledge base in various domain applications - LLM-GenFramework. The LLM-GenFramework will combine all aspects of knowledge base and fact base creation and updating, the creation of specialised data for LLM models (embeddings), and allow for contextual data retrieval and data organisation and summarisation.
The following innovation features will be developed:
- The contextual search and identification of the locations of a given element with dependencies to other elements.
The LLM-GenFramework module will have the ability to perform contextual search and identify the locations of the searched element and the dependencies of a given element on other elements. It will be able to identify dependencies between elements, together with the type of dependency and situational context and achieve at least 95% of correctness in search and identification dependencies. - Accurate positioning of information search effects.
The LLM-GenFramework module will have the functionality to correctly prioritise search results so that the most relevant answers for a given query are shown first. The relevance of the answers should depend on the context of the query. In addition, the order of presentation should be enriched with justification - why a given result is presented. - Quality and correctness of search descriptions.
The LLM-GenFramework module will have the ability to generate search result descriptions. The descriptions will contain the most important information from the search results in the context of a given query, indicating the relevance of a given result and the necessary actions to be carried out. Metrics provided in the Technical section. - Automatic building of an automated, quantitative investment strategy knowledge base.
The LLM-Invest product will enable automatic retrieval of automated, quantitative investment strategy data including source code, documentation, test cases, results of conducted tests on historical data for different instruments, results of actual investment using the strategy for different instruments and others. The database will be used for contextual search and enriching queries to LLM models with additional contextual information.
Two new products will be developed:
- LLM-GenFramework - a generic framework using RAG techniques to correctly prepare queries for LLM models to avoid hallucination effects and improve the precision of the answers. The product will be used by Holisun to offer LLM solutions based on RAG techniques for various industries, being integrated into existing Holisun applications.
- LLM-Invest - a specialised product based on the LLM-GenFramework designed to create, analyse and develop automated, quantitative investment strategies. The product will be used by AI Investments and inbestMe and will be integrated into both companies' systems.
Both products represent an entry into existing markets - IT applications and the investment market with innovative products based on state-of-the-art LLM models trained and re-trained using RAG methods.
Funding programme: EUROSTARS 3 Call 7
Project duration:
24 months (01/07/2025 - 30/06/2027)
Budget:
- Total budget: 1.6M€
- Holisun budget: 400k€
Steps and reports:
- Initiation, organization and preparation of research and data infrastructure -
- Advanced research and development of the LLM-GenFramework architecture -
- Integration, final testing and validation in real applications 06/12/2026 - 30/06/2027
Partners:
