Vision and positioning

The cloud means paying to use someone else's computer

Read that sentence again. On its own it sums up the trap most companies walk into when they adopt AI today.

Data centres are not invulnerable. They burn, they fail, they get attacked. In 2024 the serious incidents ran into the dozens, and every minute of downtime costs thousands of euros to companies with no fallback. Your data, your models, your intellectual property: all of it sits on servers you do not control, in jurisdictions you did not choose.

Meanwhile the large technology firms are borrowing heavily, some 700 billion dollars, to build ever more data centres. Who pays that bill? You do, through your cloud subscriptions. And on top of that, your personal and business data feeds those same firms' proprietary AI models. You are paying to train the competition.

The BrainDup story: from vision to framework

Stage 1: the vision

BrainDup started from a simple observation: sovereign AI is technically possible, but there was no framework to build it with.

The parts exist. Ollama runs language models locally. Milvus stores and searches vector embeddings at scale. Neo4j structures knowledge into relational graphs. n8n orchestrates the flows. Nextcloud handles files and collaboration. Each of those technologies is mature, open source and deployable on premises.

The pivotal article came in September 2024, on the Neo4j blog: “GraphRAG Agent with Neo4j & Milvus”. It laid the foundation of what became BrainDup's technical signature: combining semantic vector search (Milvus) with a structured knowledge graph (Neo4j) inside a single agent.

That approach, GraphRAG, has been validated scientifically by CETIC, the Belgian centre of excellence in information and communication technology. Measured result: GraphRAG reaches 81.67 % accuracy, against 57.50 % for vector-only methods. Twenty-four points gained.

Stage 2: proof in the field

A vision is worth nothing without execution. BrainDup matured through real projects in different domains, each one pushing the framework further.

• MethodMIND: 2,000 PubMed abstracts indexed in two hours with vector RAG, using SciBERT and T5 reranking.
• PST2RAG: a pipeline from PST to MBOX to Markdown to embeddings and a Neo4j graph. This is where the dual schema was born.
• IA-ACTES, in industrial production: a deed that took seven hours of work is produced in one, 100 % accuracy on commercial leases, 22 pharmacies validated in a batch. Built in four months, with a productivity factor of 2.9.

Stage 3: the framework today

BrainDup took a decisive step by bringing Claude Code and the BMAD method (Business, Market, Architecture, Delivery) into its enterprise edition.

You no longer start from a blank page. You no longer put up with a ready-made app. You start with a complete development environment, working and extensible, shaped for a given line of business. The methodology is in place. The development rules are in place. The developer is not left alone in front of a new framework: the community backs them, and personal guidance helps build sovereign, modern solutions in record time. No stress.