AI research applied toreal-world problems.
Not for other researchers — for those who need to understand, invest, and act. Ask our research corpus directly.
Publications
Follow our research on ResearchGate ↗MAY 2026
Conference Paper · NLPAICS 2026
Media Analysis & Social Signals
Bloc-Conditional Event States: Measuring Cross-Coverage Divergence for Threat-Intelligence Analysis
Maryam Fooladi, Federico Bottino, Alberto Trivero
A content-level method for open-source threat intelligence: each editorially-coherent outlet bloc is represented as a density matrix on a 15-dimensional framing-feature space, and cross-bloc divergence is measured via trace distance. Across two contested events, the gap between state-aligned and mainstream-Western coverage exceeds US-right/US-left polarization by roughly 1.8× — with eigenvector decomposition attributing the divergence to interpretable framing axes (economic consequences on Hormuz, morality on Navalny).
MAR 2026 (APR 2026 UPDATE)
Research Paper · Preprint
Epistemic Knowledge Access
Retrieval Is Not Enough: Why Organizational AI Needs Epistemic Infrastructure
Federico Bottino, Carlo Ferrero, Nicholas Dosio, Pierfrancesco Beneventano
A framework that gives every unit of organisational knowledge a computable epistemic status — typed, scored, dynamically maintained. The substrate your AI agents need to reason reliably, avoid contradiction, and scale institutional memory.
MAR 2026
Technical Report
Autonomous Scientific Discovery
MAGELLAN: Autonomous Cross-Disciplinary Scientific Discovery via Multi-Agent Adversarial Scrutiny
Alberto Trivero
A 15-agent open-source system that reads across scientific silos, connects existing knowledge into testable hypotheses, and subjects them to adversarial scrutiny — killing 86% of its own ideas. Includes two computationally verified predictions and a 42-page arXiv paper documenting 22 sessions and 273 hypotheses.
FEB 2026
Conference Paper · LREC 2026
Media Analysis & Social Signals
A Multi-Layer AI Framework for Information Landscape Analysis
Maryam Fooladi, Federico Bottino
A framework proposing a multi-layer AI architecture for information landscape analysis — integrating strategic reasoning, operational automation, and human-AI collaboration across the full lifecycle of new ventures.
FEB 2026
Conference Paper · LREC 2026
Media Analysis & Social Signals
Beyond Sentiment: Why Traditional NLP Fails Political News — and How LLMs Can Bridge the Gap
Maryam Fooladi, Federico Bottino
A comparative study identifying "neutral collapse" — the systematic failure of sentiment models to capture the rhetorical richness of political discourse — and demonstrating how LLMs can bridge the gap between surface-level polarity and deep narrative structure.
NOV 2025
Policy Paper
AI Compliance & AI Law
Assessing AI Act Compliance: An LLM Tool for Enterprises
Giacomo Conti, Alberto Trivero
ShikAI's first policy paper maps the combined regulatory surface of EU AI Act, GDPR, DSA and DMA — and presents an LLM-based tool that helps enterprises assess their compliance posture without sacrificing innovation.
“We are entering an era of associative intelligence — biological and artificial. Yet the driving force remains unchanged: human imagination. AI is not just computer science; it is applied philosophy, grounded in mathematics and logic. This is the era of philosophers and machines.”
Federico Bottino
CEO, Kakashi Venture Accelerator
Scientific Committee
Our research is reviewed and guided by prestigious academics and researchers.
Current Investigations
AI Compliance & AI Law
Regulation as competitive advantage, not a constraint.
Epistemic Knowledge Access
How we access knowledge — and how AI changes the rules.
Media Analysis & Social Signals
Narrative, framing and social signals — reading what media mean, not just their tone.
Autonomous Scientific Discovery
AI that reads across silos, finds what humans miss, and kills its own bad ideas.
LLMs Performance Benchmarking
Independent benchmarks. No sponsors. No convenient rankings.
Neurocognitive Architecture & Organizational Psychometrics
How minds organize, decide, and adapt — measured, not assumed.
Datasets
The corpora and evaluation suites behind our papers, released openly so the work can be reproduced and built upon.
OIDA Evaluation Corpora
Benchmark datasets and evaluation harnesses for the OIDA epistemic knowledge framework.
View on GitHub ↗More datasets are on the way. All releases live on github.com/kakashi-ventures.
Commitments
What we promise about how we do research — not after the fact, but going in. These are not marketing values; they are constraints we accept upfront and that you can hold us to.
01
Open by default
We publish our research openly. Code, data, methodology, and findings, when not bound by partner confidentiality, go public. Where we use proprietary data, we document the methodology so others can reproduce the work on their own.
02
Reproducibility over polish
A paper that cannot be reproduced is a press release. We commit to releasing the artefacts — datasets, scripts, evaluation harnesses — required to verify our claims.
03
Independent peer review
Our research is reviewed by an external Scientific Committee of academics and researchers from MIT, University of Turin, Politecnico di Torino, and GraphAware. We publish our committee composition and rotate it.
04
European AI sovereignty
We work primarily with European researchers, European institutions, European data, and EU regulatory frameworks — the AI Act, GDPR, DSA, and DMA. We contribute to building a European AI research stack.
05
Researcher independence
Researchers we work with retain authorship and freedom to publish, including findings inconvenient for KVA. We do not sign NDAs that bind academic publication.
06
Funding transparency
We declare funding sources, industrial partnerships, and conflicts of interest on every paper.
Code on GitHub
Datasets, evaluation harnesses, and reference implementations behind our papers. Open by default — fork it, reproduce it, build on it.
github.com/kakashi-venturesReproducible by design
- → Public datasets & evaluation harnesses
- → Reference implementations for each paper
- → Documented methodology for proprietary data
Contribute Hub
Stay Informed
Receive new papers and analysis.
No spam — only research. One update when we publish something new.