🤝 Top Collaborators
.png&w=640&q=75)
🤓 Latest Submissions
Meridian - Supply Chain ESG Compliance Agent
Meridian is an autonomous supply chain ESG compliance intelligence platform that monitors the entire public web to detect human rights, environmental, and financial violation signals at supplier companies 4 to 8 weeks before they escalate into scandals or regulatory enforcement. It is built for three high-penalty regulations: the EU CSDDD (up to 5% of global turnover), the US UFLPA (import bans), and Germany's LkSG (up to 2% of turnover). Three innovations power the platform. (1) Geo-Native Signal Harvesting: a LangGraph multi-agent engine crawls 60+ multilingual sources like Mandarin worker forums, regional NGO reports, government portals using local-country IPs via Bright Data Residential Proxies, SERP API, Web Unlocker, Scraping Browser, and the MCP Server. (2) Violation Velocity Scoring: every harvested signal is enriched by the AI/ML API intelligence layer (category, severity, sentiment, summary), then scored 0–100 across four risk stages. (3) Regulatory Mandate Mapper: signals become submission-ready reports in each regulator's exact format. Meridian also offers voice-enabled monitoring: Speechmatics transcribes audio evidence (worker testimony, press conferences, broadcasts), and the AI/ML API turns transcripts into structured compliance signals feeding the same scoring pipeline. The stack is a 3-service architecture: Next.js 15 frontend, a Hono/TypeScript API, and a Python FastAPI + LangGraph AI backend, deployable to Vercel and Railway/Render with Supabase, Upstash, and Qdrant Cloud.
31 May 2026
SOCsentinel: Multi-Agent LLM SOC Analyst
SOCsentinel is a fully autonomous multi-agent LLM platform that replicates the entire SOC (Security Operations Center) analyst hierarchy using 9 specialized AI agents running Qwen3 models on AMD MI300X via vLLM and ROCm. Security Operations Centers are overwhelmed: 11,000 alerts/day, 45-minute average triage time, 78% analyst burnout, and 68% of alerts go uninvestigated. Human analysts simply cannot keep up with modern threat volumes. SOCsentinel solves this by deploying a coordinated multi-agent pipeline that automates L1-L3 triage, evidence collection, MITRE ATT&CK mapping (RAG-grounded with 697 techniques via ChromaDB), Sigma rule generation, investigation reporting, and containment playbook creation - all in under 2 minutes. Each of the 9 agents uses a tailored Qwen3 model (4B for fast triage, 7B for analysis, 14B for report writing) with role-specific system prompts and guardrails. The platform features real-time SSE streaming where users can watch the entire investigation pipeline execute live with animated agent collaboration graph visualization. A self-improving feedback loop ensures analyst corrections automatically calibrate future AI triage classifications with visible learning metrics. The Human-in-the-Loop Analyst Workbench provides confidence override sliders, auto-generated risk summaries, decision history, and full audit trail - ensuring AI augments rather than replaces human judgment. Additional features include Qwen3 "Thinking Mode" for Chain-of-Thought reasoning visibility, a one-click benchmark dashboard across 5 attack scenarios, and executive summary banners with kill chain progress and recommended actions. The tech stack consists of FastAPI + LangChain + ChromaDB on the backend, React + TypeScript + TailwindCSS on the frontend, and Qwen3 (4B/7B/14B) served via vLLM on AMD MI300X (192GB HBM3) using AMD Developer Cloud and ROCm 6.x.
10 May 2026
