HIVEMIND: Human-centred collaborative multi-agent framework for accelerating software Development and maintenance

HIVEMIND is an innovative project designed to enhance software engineering by integrating responsible, human-centric methods, tools, and best practices that capitalize on AI and data technologies. The project aspires to expedite every phase of the software development lifecycle.

 

At the core of HIVEMIND is the creation of an adaptive Large Language Model (LLM)-based multi-agent framework. This framework is tailored to foster collaboration between human developers and a constellation of AI agents. Each agent within this system is meticulously crafted to reflect and support the diverse roles found in a traditional software development team, providing specialized assistance.

 

These AI agents are fine-tuned and customized for specific tasks through various methods: fine-tuning with organizational data for personalized learning, prompt engineering for targeted responses, Retrieval Augmented Generation for enhanced knowledge retrieval, and Human-in-the-loop Machine Learning for iterative improvement involving user feedback.

 

HIVEMIND aims to develop mechanisms that streamline smart system specification. These mechanisms will have the capability to automatically generate complex requirements and facilitate agile modeling. By handling inconsistencies and ambiguities, the project aims to minimize the need for later modifications in the development process.

 

Moreover, HIVEMIND seeks to bolster design-by-contract programming across all levels of integration. This is achieved by enhancing the context awareness of AI agents involved in code development, analysis, verification, and testing. These agents will be able to access relevant documentation seamlessly throughout the development process.

Our main tasks

• Coordination and management of the project
• Initiating the requirements phase with thorough planning to capture user needs and limitations
• Adhering to a secure and quality-focused software development process
• Ensuring user data privacy and security, with foresight into user demands
• Fine-tuning Large Language Models (LLMs) efficiently with specific organizational data
• Enhancing Language Models with context sensitivity using Retrieval-Augmented Generation (RAG)
• Incorporating diverse human feedback into AI systems using LLMs
• Integrating and evaluating open-source LLM customization techniques

Partners

DFKI AI | FRAUNHOFER | HAVELSAN | IDENER.AI | MÄLARDALENS UNIVERSITET | MTU AUSTRALO ALPHA LAB | RESEARCH INSTITUTES OF SWEDEN | SIMAVI | SUPSI | THE QUEEN'S UNIVERSITY OF BELFAST | TIGA SAĞLIK TEKNOLOJILERI | UNIVERSIDAD DE ALICANTE | UNIVERSITAT POLITECNICA DE CATALUNYA

Start date – finish date

01 / 2025 - 12 / 2027