# Subhash Dasyam

> Practical technical articles about agentic AI security, MCP, LLM inference, RAG, vector database security, containers, and Kubernetes.

The canonical HTML articles are the source of truth. This file is a curated discovery map, not a duplicate content inventory.

## Start here

- [All technical articles](https://www.subhashdasyam.com/p/articles.html): Browse the complete article collection by topic.
- [About Subhash Dasyam](https://www.subhashdasyam.com/p/about.html): Author background and public profiles.

## Agentic AI and MCP security

- [Enterprise MCP Security Architecture: A Production Guide](https://www.subhashdasyam.com/2025/09/building-enterprise-mcp-architecture.html): Resource-bound authorization, tool policy, isolation, approvals, audit evidence, and negative tests.
- [Securing Agentic AI: Architecture, Patterns, and Governance](https://www.subhashdasyam.com/2025/12/securing-agentic-ai-architecture.html): Entry point for the ten-part agentic AI security series.
- [Multi-Agent AI Security: Trust Boundaries and Handoffs](https://www.subhashdasyam.com/2025/12/securing-agentic-ai-architecture_64.html): Delegation, authenticated handoffs, isolation, approval, and observability.
- [Multi-Turn Jailbreaks and Psychological Manipulation Attacks](https://www.subhashdasyam.com/2025/12/multi-turn-ai-jailbreaks-and-psychological-manipulation-attacks.html): Why single-turn defenses fail and what to test instead.

## RAG and vector security

- [Top 10 Security Issues for Vector Databases and AI Systems](https://www.subhashdasyam.com/2024/09/top-10-security-issues-for-vector.html): Practical risks and controls for vector systems.
- [Vector Injection and Vector Database Vulnerabilities](https://www.subhashdasyam.com/2024/09/vector-venom-vector-injection-how-ai.html): How poisoned vectors can influence retrieval.
- [Privacy Preserving RAG with Homomorphic Encryption](https://www.subhashdasyam.com/2025/11/building-privacy-preserving-rag-with.html): Privacy boundaries and implementation considerations for encrypted retrieval.

## LLM systems and inference

- [KV Cache Security for Long-Context LLM Inference](https://www.subhashdasyam.com/2025/12/long-context-inference-security-kv.html): Isolation, reuse, eviction, observability, and failure controls for KV caches.
- [Complete Guide to LLM Inference Servers](https://www.subhashdasyam.com/2025/05/complete-guide-to-llm-inference-servers.html): From model loading to production serving.
- [Decoding FP32, FP16, FP8, INT8 and INT4](https://www.subhashdasyam.com/2025/08/decoding-fp32-fp16-fp8-int8-int4-master.html): A simple guide to precision formats and safe evaluation.

## Containers and Kubernetes

- [Policy as Code for AI Workloads in Kubernetes](https://www.subhashdasyam.com/2025/12/policy-as-code-for-ai-workloads-in-kubernetes.html.html): Kyverno and OPA patterns for model and data safety.
- [Container SBOMs and Receipts](https://www.subhashdasyam.com/2025/05/container-receipts-missing-ingredient.html): Supply chain evidence for container images.
- [Linux Fundamentals Behind Containers](https://www.subhashdasyam.com/2025/05/linux-fundamentals-behind-containers.html): The Linux primitives that make containers work.

## Original field work

- [One Cable, 200 Gbit/s: Connecting Two GB10 Systems with ConnectX-7](https://www.subhashdasyam.com/2026/07/one-cable-200-gbits-connecting-two-gb10.html): Reproducible direct-link setup notes and measured throughput.
- [Damn Vulnerable AI Bank](https://www.subhashdasyam.com/2025/12/damn-vulnerable-ai-bank.html): An intentionally vulnerable AI banking project for defensive learning.

## Discovery feeds

- [Post sitemap](https://www.subhashdasyam.com/sitemap.xml): Canonical post discovery.
- [Page sitemap](https://www.subhashdasyam.com/sitemap-pages.xml): Canonical static page discovery.
- [Atom feed](https://www.subhashdasyam.com/feeds/posts/default): Latest and updated posts.