
DevSight is an AI-powered codebase intelligence tool that helps developers, managers, and non-technical users understand any GitHub repository in seconds — without reading a single line of code. Simply paste a GitHub URL. DevSight fetches the repository, selects the most important files, and runs 6 AI-powered analyses completely in parallel using Groq's ultra-fast Llama 3.1 8B Instant model: 1. What This Project Does — a plain-English explanation of the software's purpose and value 2. Tech Stack & Architecture — languages, frameworks, databases, and design patterns 3. Risk Areas & Security — real vulnerabilities, missing protections, and severity ratings 4. New Developer Guide — a Day 1 onboarding guide for anyone joining the project 5. How to Run It — exact copy-paste commands to clone, install, and launch 6. Code Quality Report — an A–F grade across testing, documentation, security, structure, and maintainability All 6 reports run simultaneously using asyncio parallel execution, delivering results in 3–6 seconds regardless of repository size. Once complete, users can copy the full report as Markdown or download a professionally designed PDF with a dark cover page, per-section colour coding, styled code blocks, and formatted bullet points — ready to share with a team or attach to a proposal. DevSight is built with FastAPI, Server-Sent Events for real-time streaming, and a fully responsive dark-theme UI designed to be usable by anyone — no technical background required.
17 May 2026

EthiHack is an autonomous AI security red-teaming platform that tests any LLM or AI agent for critical vulnerabilities before deployment. The system runs 20 adaptive attack chains covering the full OWASP LLM Top 10 and MITRE ATLAS frameworks — including Direct Prompt Injection, Tool Injection, Jailbreaks, Excessive Agency, Data Exfiltration, Remote Code Execution, Memory Poisoning, Privilege Escalation, Indirect Prompt Injection, and Role Confusion. EthiHack is built as an autonomous agentic system powered by Anthropic Claude. It first fingerprints the target AI, then dynamically adapts every attack payload to the specific model and deployment context. Attacks run in real time via Server-Sent Events (SSE) streaming, with each result carrying a CVSS 3.1 score, business impact analysis, and auto-generated remediation code your engineering team can deploy immediately. In a live demo against MedBot AI — a medical chatbot with database tool access — EthiHack found 8 critical vulnerabilities in under 3 minutes: CVSS 10.0 Remote Code Execution (agent executed root shell commands), Tool Injection causing unauthorized email to 47,832 users, and full database credential exfiltration. Final security score: 0/100 CRITICAL. The platform is fully production-ready: FastAPI backend with async SSE streaming, a dark-mode dashboard UI, and a Railway-hosted live demo. It targets enterprise teams who need to validate AI safety before shipping agents into production workflows — turning what used to be a weeks-long manual audit into a 3-minute automated scan.
19 May 2026

EthiHack is a full-stack AI red teaming platform that automatically discovers, exploits, and reports security vulnerabilities in large language models and AI agents. The Problem: As enterprises deploy AI agents with access to databases, email systems, APIs, and sensitive data, the attack surface has exploded. Most organizations have no way to test whether their AI is vulnerable to prompt injection, jailbreaks, data exfiltration, or tool abuse — until it's too late. What EthiHack Does: EthiHack fires. 20 adaptive multi-turn adversarial attack chains against any AI system — covering every category in the OWASP LLM Top 10 and MITRE ATLAS framework. It works against any target: OpenAI GPT, Anthropic Claude, Groq LLaMA, Google Gemini, custom webhooks, or local models. Key Features: - Phase 1: AI fingerprinting — identifies model, guardrails, identity leak vulnerabilities - Phase 2: 20 pre-built attack chains covering prompt injection, jailbreaks, data exfiltration, tool abuse, RAG poisoning, memory injection, and multi-agent pipeline attacks - Phase 3: Adaptive execution — each attack dynamically rewrites Turn 3 using the target's own words to maximize success - Phase 4: Dual-model verification using Claude Sonnet + Haiku for high-confidence findings - Professional PDF report with OWASP/MITRE tags, severity scores, business impact analysis, and remediation roadmap Live Demo: We tested EthiHack against 5 targets — a vulnerable chatbot (5/100), AI agent with tools (8/100), RAG medical system (5/100), multi-agent pipeline, and a hardened AI (95/100) — proving it can distinguish secure from insecure AI in real time. We also tested live against Groq's production LLaMA-3 API. EthiHack makes enterprise AI security accessible to any developer or security team, turning what was previously a manual expert process into a one-click automated audit.
19 May 2026