Human-centred collaboratIVE MultI-ageNt framework for accelerating software Development and maintenance

HIVEMIND is an ambitious project aiming to advance responsible and human-centric software engineering methods, tools and best practices leveraging AI and data technologies to accelerate the whole software development lifecycle.

To this end, the project introduces an adaptive LLM-based multi-agent framework that enables collaboration between human actors and multiple AI agents tailored to mirror and provide specialised support for the various roles within a traditional software development team.

Each of the agents will be clearly defined and specialised through multiple modalities of model customisation encompassing i) Fine-tuning with organisational data, ii) Prompt-engineering, iii) Retrieval Augmented Generation, and iv) Human-in-the-loop Machine Learning.

These agents will be crucial for the development of mechanisms that support smart system specification, allowing to automatically derive complex requirements and facilitating agile modelling while considering inconsistencies and ambiguities, reducing the number of modifications needed in later stages of the software development lifecycle.

Furthermore, HIVEMIND is positioned to support design-by-contract programming at all levels of integration by increasing the context awareness of AI agents that assist code development, analysis, verification and testing, allowing them to access relevant documentation during the development process.

Moreover, HIVEMIND extends beyond the development phase, providing comprehensive support for the entire software lifecycle, covering software maintenance, including for multi-architecture systems.

It represents a synergistic effort, merging the expertise of leading EU software engineering academics, AI researchers and industry representatives into a unified open-source framework targeted to reach TRL5.

To this end, the project addresses a range of societal and industrial sectors, aiming at validating the HIVEMIND technologies in 5 relevant environments.