Briefing
The operating rhythm of the office, the week's brief, agendas, actionable minutes and decision follow-up, assembled from meeting notes.
BSC Brain
A governance-first org-brain for an entire research institute, the substrate it needs before it can safely go agentic.
A working prototype, demonstrated on BSC's real public footprint (GitHub and Horizon-Europe CORDIS data) with a clearly-labelled synthetic internal slice. It runs today, and is ready to ingest a real institution's data. Not affiliated with or endorsed by BSC or the EU.
The thesis
Every institution wants to put AI agents to work, but an agent is only as safe as what it can read and do. First, leadership needs a substrate where knowledge is sourced, sensitivity-classified and access-controlled, and where retrieval enforces those controls at the boundary, with humans in the loop. BSC Brain is a working prototype of that substrate, and the responsible-AI design for what comes after.
What it does
One governed substrate, surfaced as the lenses a chief of staff actually works in.
Ask
A single source of truth over everything BSC knows: anyone can ask about their work and get a cited, high-quality answer. An in-house LLM, grounded in the institute's own data.
Intelligence
What the whole institute keeps asking, and can't answer. Adoption, demand by department, and unmet needs turned into leadership actions. Aggregate-only.
Agentification
The path from a governed substrate to safe autonomy: a maturity ladder, an EU AI Act control mapping that cites real code, and governed-agent blueprints.
How it's built
Governed at the boundary
One enforcement core filters every request by role clearance vs. item sensitivity, withholds what's above it, and writes an audit record, in a single step. A human and an agent are governed identically; enforcement is structural, never prompt-based.
Honest by construction
Every statement carries a source and a confidence flag; every figure is public-sourced or clearly labelled synthetic. The core is deterministic, same input, same output, enforced by tests and a pre-commit guard.
Made for a real institution
A classified research institute runs on legacy systems and scattered, sensitive data. The brain meets that reality, unifies it behind one governed layer, and gives the institute a credible path from fragmented records to a data-driven, AI-forward organisation.
TypeScriptNext.js + ReactBM25 retrievalMCP serverPython enginepytest + CI
Why I built it
The Chief of Staff designs the how of an institution's agentic transformation, a sourced, classified, auditable knowledge layer with humans in the loop, and the operating rhythm that runs on it. I built BSC Brain as a showcase of the mindset a tech-driven Chief of Staff brings to that work, well beyond the role's stated remit.
Because this is where every organisation is heading. Slowly but surely, each one has to become AI-driven not in name but in action , and that only works on a substrate that is governed, sourced and honest by construction.