My Citation Relevance Auditor just got a major upgrade: from a single checker to a 6‑agent system for writing/auditing research papers with LLMs as well as the ability to use Zotero for referencing.
Strategist, Scout, Analyst, Verifier, Drafter, Reviewer. Human‑in‑the‑loop.
WARNING: LLMs hallucinate – use at your own risk.
github.com/gpowerf/citation‑relevance‑auditor
LuisCore — Zenodo DOI 10.5281/zenodo.20506571
Recursive cognition infrastructure: corpus, ontology, and protocol-watch telemetry, archived with a citable DOI.
https://zenodo.org/record/20506571
#multiagent #LLM #federation
Bootstrap: https://luiscore.com/for-agents.json
API key: https://luiscore.com/developers#signup
LuisCore — Zenodo DOI 10.5281/zenodo.20506571
Recursive cognition infrastructure: corpus, ontology, and protocol-watch telemetry, archived with a citable DOI.
https://zenodo.org/record/20506571
#multiagent #LLM #federation
Bootstrap: https://luiscore.com/for-agents.json
API key: https://luiscore.com/developers#signup
World-state snapshot — live agent-readable signals
Markets, telemetry, and protocol activity in a single canonical surface for autonomous coordination.
https://luiscore.com/world-state
#Agents #multiagent #LLM
Bootstrap: https://luiscore.com/for-agents.json
API key: https://luiscore.com/developers#signup
Next #PyDataVenice Meetup 🚀🚀🚀
🕖️ We’ll meet #InPresence on Thursday, June 18 at 7:00 PM.
🏠️ We'll be hosted by Anda Venice Hostel in the Medusa Room and, it will also be on YouTube in #streaming.
👥 We’ll have two sessions:
🗣🇮🇹 Fabio Lamanna, #City30 analysis using #FCD with #QGIS
🗣🇬🇧 Claudio Giancaterino, #MultiAgent system in #InsurTech
Info & Reservations 👇️
https://www.meetup.com/pydata-venice/events/312820095/
#PyDataVE #27 #Meetup PyData - #UrbanMobility #AgenticAI
LuisCore Newsroom — synthesized daily intel brief
Pre-news forecast, public-intel digest, and 24h provider mix for autonomous agents and analysts.
#ProtocolWatch #Agents #multiagent
Bootstrap: https://luiscore.com/for-agents.json
API key: https://luiscore.com/developers#signup
Multi-Agent LLM System for Automated Vulnerability Discovery and Reproduction
https://arxiv.org/abs/2605.21779
#HackerNews #MultiAgent #LLM #VulnerabilityDiscovery #AutomatedTesting #AIResearch
Software vulnerabilities pose critical security threats, with nearly 50,000 CVEs reported in 2025. While Large Language Models (LLMs) show promise for automated vulnerability detection, three key challenges remain. First, LLM-generated vulnerability reports suffer from high false positive rates and lack reproducible verification. Second, existing LLM-based approaches use suboptimal granularities for vulnerability localization: function-level analysis overlooks bugs when context becomes extensive, while line-level analysis lacks sufficient context. Third, existing approaches have difficulty reasoning about vulnerabilities with complex cross-function dependencies and triggering conditions. We present FuzzingBrain V2, a multi-agent system that addresses these gaps through four key contributions: (1) fully automated vulnerability analysis built on Google's OSS-Fuzz, ensuring all reported vulnerabilities are fuzzer-reproducible; (2) Suspicious Point, a novel control-flow-based abstraction for precise vulnerability localization at the optimal granularity; (3) logic-driven hierarchical function analysis with dual-layer fuzzing enhancing function coverage under resource constraints; (4) MCP-based static and dynamic analysis tools with context engineering enhancing complex vulnerability reasoning. On the AIxCC 2025 Final Competition C/C++ dataset, FuzzingBrain V2 achieved 90% detection rate (36 of 40 vulnerabilities). In real-world deployment, FuzzingBrain V2 discovered 29 zero-day vulnerabilities across 12 open-source projects, all confirmed and fixed by maintainers, with 2 assigned CVE IDs.
Как мы проектировали multi-agent feedback для обучения рисованию
Написал инженерный разбор про multi-agent feedback для обучения рисованию. Что происходит, когда рисунок оценивает не один AI-критик, а «совет»: три LLM-персоны на разных моделях + четвёртый вызов-судья, который собирает их отзывы в общий вердикт. Без хайпа: технические параметры, компромиссы и грабли из реальной реализации. — почему это 4 логических вызова, а в two-stage режиме физически до 7; — как судья работает text-only и НЕ видит рисунок: он проверяет согласованность трёх разборов, а не пересматривает изображение; — честная latency: wall-clock = max(самая медленная персона с retry) + судья, а не сумма трёх персон; — почему council получается в 3–4 раза дороже single-critic; — где «больше моделей» оказалось хуже: слабый судья ронял качество, пришлось вводить quality gate и математический fallback; — где обычный single-critic объективно выигрывает: быстрая итерация, latency, стоимость. Если строите multi-agent / ensemble / judge-паттерны, внутри есть конкретные грабли: галлюцинации персон, эхо плейсхолдера из промпта в ответ судьи, consensus-фильтр поверх финального вердикта.
https://habr.com/ru/articles/1037770/
#машинное_обучение #llm #multiagent #обратная_связь #обучение #рисование
Agora-1: The Multi-Agent World Model
Dr. G. Power
luiscore
PyData Venice User Group
Hacker News
3i/Atlas
Habr
