🤓 Latest Submissions
Blitz TraceGrid
Blitz TraceGrid is visual execution intelligence for AI-generated software. The problem is simple: AI can create code faster than teams can understand its behavior. Developers and security reviewers still need to know what happens after a user action, where trust boundaries appear, which path deserves investigation, and whether the visible behavior is well-supported or just a weak inference. TraceGrid starts from uploaded source files, a ZIP archive, a repository path, public GitHub URL, or deterministic demo fixture. It extracts AST-lite evidence from Python and TypeScript/JavaScript, reconstructs a causal graph across UI, API, backend, security, dependency, and data layers, then lets the user choose a Trace Target. That target becomes a focused trace slice with incoming context, outgoing branches, and one primary replay path selected for animation. The new Behavioral Trust Map scores AI-generated behavior without adding more chatbot layers. It highlights behavior confidence, unknown zones, architecture drift, execution entropy, security consequence hotspots, and hallucination-like signals that need confirmation. In plain language, it asks: can we trust this path yet, or do we need runtime proof? The flagship Run AI Investigation flow makes the app agentic. TraceGrid scores graph targets by risk signal, connectivity, and layer priority, selects a high-value investigation root, traces it, runs Architecture, Security, Execution, and Explainer agents, and outputs root cause, attack surface, replay path, recommended fix, confidence, and a plain-English explanation. TraceGrid is deployed as a production-shaped web app with Next.js, FastAPI, Docker, Nginx, Vultr, Featherless inference, and Speechmatics voice control. Current build: static causal reconstruction, not full runtime telemetry. The roadmap is OpenTelemetry, browser/backend hooks, eBPF event capture, taint/data-flow analysis, causal diffing, and deterministic replay.
19 May 2026
Blitz SOC
Blitz SOC is an explainable hybrid security operations workflow designed for the AMD Developer Hackathon under the AI Agents & Agentic Workflows track. The platform transforms synthetic security telemetry into correlated, analyst-ready incident intelligence using three execution modes: • Baseline Mode — deterministic control-path analysis using fixed logic • Local SOC Mode — resilient local orchestration with state snapshots, correlation, and approval workflows • CrewAI Mode — AI-assisted analyst reasoning powered by CrewAI and Qwen inference served through AMD ROCm and vLLM Unlike traditional SOC systems that rely entirely on static rules or opaque AI automation, Blitz SOC separates AI-assisted reasoning from deterministic governance. CrewAI coordinates analyst-style workflows such as classification, enrichment, escalation reasoning, and response preparation. Qwen provides analyst-facing contextual reasoning, incident summarization, and response generation. AMD Developer Cloud powers the accelerated inference environment through ROCm and vLLM. After AI enrichment occurs, Blitz applies deterministic governance layers including: • Correlation • Risk scoring • Escalation policy • Approval gates • Integration routing • Incident export workflows This separation keeps the workflow explainable, auditable, resilient, and operationally safer than fully autonomous approaches. The project includes: • Explainable execution traces • Real-time incident streaming • AI reasoning visibility • SOC response timelines • Human approval gates • Production integration mapping • AMD inference proof panels • Multi-mode fallback resilience Blitz SOC demonstrates how open-source AI models and agentic workflows can enhance SOC analyst efficiency while preserving deterministic governance and operational safety.
10 May 2026
