Founder & CEO: Aj Solution | Prime property partners | Engineering Students

youtube

devto

github

Abhishek Jha  • Peerlist

AutoBench: Production-Grade Prompt Benchmarking &amp; SecOps ShieldAutoBench is an enterprise-grade, real-time prompt profiling, optimization, and security isolation platform. It evaluates, sanitizes, and distributes LLM prompts using concurrent thread sandboxing, automated red-teaming, live shadow routing split proxies, and multi-region edge mesh latency analysis.Live Demo: <a target="_blank" rel="nofollow" class="break-all link-break text-gray-500 hover:underline" href="https://autobench-realtime.onrender.com/">Deployed on Render</a> (may take ~30s to cold-start on the free tier)🚀 Key Features<ul><li>📦 Concurrent Sandbox Thread Streamer: Spawns isolated sandbox worktrees in parallel (Sandbox A: Baseline, Sandbox B: Stress/Adversarial, Sandbox C: Codex-Optimized) streaming real-time logs over a bi-directional WebSocket tunnel.</li><li>🛡️ SecOps Sanitizer Shield: Active input sanitizer that intercepts prompt injections, blocks restricted system execution commands, prevents PII/Credential leakage, and enforces strict JSON output compliance.</li><li>🔀 Shadow Traffic Split Routing Proxy: Simulates a live 99/1 production traffic split with an automated LLM-as-a-Judge scoring engine and a dynamic circuit breaker that terminates candidate routing if semantic accuracy drops below a 94% safety threshold.</li><li>🌐 Global Telemetry Edge Mesh: Captures high-resolution performance, TTFT (Time-To-First-Token), and percentile latency benchmarks (P50, P95, P99) across three distributed simulated nodes (us-east-1, eu-west-1, and ap-south-1).</li><li>💀 Adversarial Red-Team Simulator: Runs an automated suite of 50 multi-vector prompt injection attacks to stress-test prompt alignment and generate an interactive Vulnerability Matrix report with actionable remediation.</li><li>🤖 CI/CD Gate Validator: A headless gate validation script verifying target performance against strict enterprise regression limits, exiting with standard exit codes (0 for Pass, 1 for Breach) to fit pipeline environments.</li><li>🖥️ Cloud IDE Workspace: In-browser code editor with Monaco-grade features including syntax highlighting, file tree navigation, terminal emulation, and real-time Git diff overlays.</li><li>💎 SaaS Subscription Tiers: Interactive billing panel with Starter, Pro, Teams, and Enterprise tiers featuring a secure sandbox checkout flow.</li><li>🔗 A2A Mesh Visualizer: Real-time Agent-to-Agent mesh network topology with live latency heatmaps and node health monitoring.</li><li>🏠 Enterprise Homepage: SEO-optimized marketing homepage with interactive modals for extensions, security trust center, legal compliance, and enterprise contact.</li></ul>🛠️ Tech Stack &amp; Architecture<h3>Frontend</h3><ul><li>Framework: React 19 + TypeScript + Vite</li><li>Styling: Tailwind CSS v4 (harmonious dark UI, HSL glow vectors, micro-animations)</li><li>Icons: Lucide React</li><li>Data Layer: Custom WebSocket stream hook (useAutoBench.ts) + Fetch API</li></ul><h3>Backend</h3><ul><li>Runtime: Node.js + ESM Modules</li><li>Framework: Express v5 + standard ws WebSocket connection manager</li><li>Simulation Engine: Promise-isolated orchestrator + performance micro-benchmarking</li></ul><h3>Deployment</h3><ul><li>Hosting: <a target="_blank" rel="noopener noreferrer" class="break-all link-break text-gray-500 hover:underline" href="http://Render.com">Render.com</a> (Web Service, free tier)</li><li>CI/CD: Auto-deploy from main branch via Render Git integration</li></ul>

AutoBench: Production-Grade Prompt Benchmarki

AI ATHARV is an advanced AI-based system designed to blend intelligence with innovation. While the full capabilities are under wraps for now, the project aims to solve real-world problems through smart automation, decision-making algorithms, and adaptive learning. It's built with a vision to push the boundaries of what AI can do — in daily life, industry, and beyond.

Ai Atharv

Project Name: Guru Cortex Project Overview Project Cortex is a personalized, proactive AI study buddy designed to fundamentally change how students interact with educational AI. Unlike most AI tutors that simply wait to answer questions, Cortex utilizes a proactive approach—it actively pays attention to the student, notices when they are confused, and offers assistance before they even have to ask.Core Problem &amp; Solution Current educational AI relies on reactive prompts. Project Cortex solves this by acting as a highly observant tutor that provides affordable, accessible, and localized learning. It achieves this through a Privacy-First Hybrid Architecture (Edge + Cloud), ensuring that sensitive vision processing happens locally on the user's device while complex reasoning is handled securely.Key Features<ul><li>Smart Vision System &amp; Emotional Check-ins: Using a webcam, the system monitors for gaze, fatigue, and frustration. It detects when a student is confused or tired and proactively suggests new approaches or study breaks.</li><li>Local Knowledge Base (DeepSearch): Instead of relying on generic internet data, Cortex indexes the student's personal PDFs and notes. This allows it to provide highly contextual answers based on the user's actual study materials.</li><li>Active Learning Mode: Features a "Teach Mode" that prompts the student to explain concepts back to the AI to verify and enhance comprehension.</li><li>Multilingual Support: Offers conversational AI capabilities in Hindi and other regional languages.</li><li>Drawing &amp; Diagrams: Automatically generates visual flowcharts to help simplify complex text.</li><li>Focus Features: Detects distraction and actively alerts the student to stay focused on their screen.</li></ul>Technology Stack &amp; Architecture The system is built on a Hybrid Edge-Cloud model to balance performance with strict privacy:<ul><li>AI Models: Powered by Quantized Google Gemma (2B/9B) for language processing and Whisper Tiny for efficient voice input and output.</li><li>Vision &amp; Emotion Processing: Utilizes MediaPipe and OpenCV locally on the device, ensuring 100% video privacy without sending visual data to the cloud.</li><li>Storage &amp; Memory: Employs ChromaDB for a Local GraphRAG vector store (to search notes) and SQLite to track student progress and spaced repetition.</li><li>Interface &amp; Orchestration: Features a Next.js dynamic dashboard with Framer Motion UI, orchestrated by a Django/FastAPI server hosted on Google Cloud Run. Mermaid.js is used for generating diagrams.</li></ul>Future Development Roadmap<ul><li>Multimodal Learning: Integrating Multimodal Gemma to analyze diagrams from textbook photos and digitize handwritten notes via OCR.</li><li>Cross-Platform App: Transitioning to Flutter for seamless desktop-to-mobile syncing and running Gemma models locally on mobile devices.</li><li>Privacy-Safe Study: Enabling secure collaborative learning between students using peer-to-peer (P2P) encryption and the ability to share Knowledge Graphs without sharing raw documents.</li></ul>

GuruCortex - AI Learning Companion

An interactive OpenEnv simulation for autonomous agents to optimize cloud infrastructure costs and resolve critical security incidents. HF Space GitHub OpenEnv 📋 Overview This simulator models a realistic cloud environment with EC2 instances, Elastic IPs, IAM users, and security alerts. Autonomous agents must act as Cloud FinOps and Security engineers to: ✅ Detect and eliminate cost waste ✅ Resolve critical security breaches ✅ Maintain production stability Perfect for: Training and evaluating LLM-based agents on real-world cloud operations scenarios.