🚀 AI Systems & Autonomous Agents: Building deterministic agent execution platforms, tool-calling pipelines, and evaluation harnesses. 🛡️ Zero-Trust Sandboxing: Isolating untrusted code execution inside ephemeral Docker containers with token-gated host-side proxies. ⚡ Open Source Engineering: Production contributor to Soup across cross-platform Win32/POSIX portability, terminal rendering, and CI stability. 🎯 Current Focus: Sub-15ms streaming proxy relays, egress boundary firewalls, and defense against model supply-chain attacks. 🎓 Education: Bachelor of Computer Applications (BCA) · 2nd Year · Indira Gandhi National Open University (IGNOU), New Delhi.

const Shubham = {
    role: "AI Systems & Defensive Cybersecurity Engineer",
    location: "Delhi, India",
    education: "BCA (2nd Year) — IGNOU, New Delhi",
    domains: [
        "Autonomous Agent Runtimes",
        "Zero-Trust Docker Sandboxing",
        "Threat Intelligence & PCAP",
        "Cross-Platform Systems (POSIX & Win32)"
    ],
    languages: {
        daily: ["Python", "TypeScript", "JavaScript"],
        systems: ["C", "Bash / Shell", "SQL"]
    },
    architecture: {
        runtimes: ["Docker Sandboxes", "Subprocess / IPC", "Server-Sent Events (SSE)"],
        ai_ml: ["PyTorch", "Hugging Face", "LoRA / PEFT", "Scikit-Learn"],
        backend: ["FastAPI", "Node.js", "BullMQ", "Redis"],
        security: ["Egress Boundaries", "Threat Modeling", "Wireshark", "OWASP Top 10"]
    },
    currentFocus: "Engineering zero-leakage LLM agent execution boundaries",
    motto: "Smallest clean change that completely solves the problem."
};

Security: Zero-Trust Docker Sandbox  |  Latency: Sub-15ms Host Proxy  |  Queue: BullMQ & Redis

• Distributed autonomous AI software engineering platform executing complex multi-file codebase tasks.
• Implements ephemeral per-user Docker sandbox containers to safely isolate untrusted shell commands and code modifications.
• Engineered a token-gated host-side LLM proxy with SSE streaming relays, enforcing strict egress network boundaries to prevent credential leakage.
• Orchestrates asynchronous tasks via BullMQ and Redis with persistent workspace volume mounts.
• Stack: TypeScript · Docker · BullMQ · Redis · Fastify · OpenCode SDK

Domain: Threat Intelligence  |  ML: Unsupervised Anomaly Detection  |  Telemetry: Network PCAP Analysis

• Predictive threat intelligence pipeline analyzing raw network traffic telemetry and server access logs.
• Applies unsupervised anomaly detection to identify unauthorized payload execution and lateral movement patterns.
• Automated feature extraction across protocol headers (TCP/IP, DNS, HTTP/S) for real-time risk scoring.
• Generates actionable defensive mitigation playbooks to prevent privilege escalation.
• Stack: Python · Scikit-Learn · FastAPI · Network Telemetry · Wireshark

Visualization: 60FPS Three.js / WebGL  |  Streaming: 10 Hz Low-Latency Telemetry

• High-throughput 3D spatial mission control platform for autonomous drone swarms and satellite constellations.
• Features a 60FPS Three.js / WebGL visualization viewport mapping multi-agent spatial trajectories.
• Low-latency 10 Hz real-time telemetry streaming pipeline with responsive ground-station telemetry sync.
• Stack: Next.js 14 · Three.js / WebGL · GSAP · TypeScript

SIH 2026: 91.4% Validation Accuracy  |  Systems: Windows CRT & POSIX CI Stability

• AgriPack AI (SIH 2026): Deep learning multi-output regression models predicting food-packaging degradation curves with 91.4% validation accuracy, served via asynchronous REST endpoints.
• Open Source Systems: Production contributor to LLM toolkits across Windows CRT command quoting, airgap validation, and multi-OS CI stabilization.
• Stack: Python · PyTorch · Scikit-Learn · FastAPI · Subprocess & IPC

_{Designed & Maintained by Shubham Kumar · AI Systems Engineering & Defensive Cybersecurity}