---
title: "Ashutosh Kumar (@askuma) | Hashnode"
description: "TPM building in public. Writing about AI guardrails - architecture, implementation, and what actually works in production. Code-first. No fluff."
url: "https://hashnode.com/@askuma"
---

@askuma

# Ashutosh Kumar

@askuma · Hyderabad, India · Joined June 2026

### [Introducing the Guardrail Platform: guardrailmesh + guardrailprobe](https://askuma.hashnode.dev/introducing-the-guardrail-platform-guardrailmesh-guardrailprobe?utm_source=hashnode&utm_medium=feed)

Architecture, live August 2026 benchmark data, and setup for guardrailmesh and guardrailprobe — open-source tools to enforce and prove AI guardrails.

10 min read · [askuma.hashnode.dev](https://askuma.hashnode.dev/introducing-the-guardrail-platform-guardrailmesh-guardrailprobe?utm_source=hashnode&utm_medium=feed)

[Aug 6](https://hashnode.com/posts/introducing-the-guardrail-platform-guardrailmesh-guardrailprobe/6a74ac6c87941d810922334f)

### [Guardrails as Code: Treating AI Safety Policies as Version-Controlled Infrastructure](https://askuma.hashnode.dev/guardrails-as-code?utm_source=hashnode&utm_medium=feed)

How policy-as-code, CI/CD regression tests, and model-agnostic enforcement turn AI safety into real infrastructure for agentic deployments.

9 min read · [askuma.hashnode.dev](https://askuma.hashnode.dev/guardrails-as-code?utm_source=hashnode&utm_medium=feed)

[Jul 27](https://hashnode.com/posts/guardrails-as-code/6a66f8c3342d63825261483a)

### [Latency Budgets vs. Security Depth: Optimizing Guardrail Middleware for Production](https://askuma.hashnode.dev/latency-budgets-vs-security-depth-optimizing-guardrail-middleware-for-production?utm_source=hashnode&utm_medium=feed)

How to design a validation pipeline that enforces rigorous compliance without making your users wait: tiered evaluation, streaming windows, and async offloading.

7 min read · [askuma.hashnode.dev](https://askuma.hashnode.dev/latency-budgets-vs-security-depth-optimizing-guardrail-middleware-for-production?utm_source=hashnode&utm_medium=feed)

[2](https://hashnode.com/posts/latency-budgets-vs-security-depth-optimizing-guardrail-middleware-for-production/6a5dcdafd4e10a40e4900e54) [Jul 20](https://hashnode.com/posts/latency-budgets-vs-security-depth-optimizing-guardrail-middleware-for-production/6a5dcdafd4e10a40e4900e54)

### [Mitigating the Three Critical LLM Vulnerabilities at the Middleware Layer](https://askuma.hashnode.dev/mitigating-the-three-critical-llm-vulnerabilities-at-the-middleware-layer?utm_source=hashnode&utm_medium=feed)

Production patterns for PII redaction, hallucination containment, and adversarial input defense, with implementation code.

6 min read · [askuma.hashnode.dev](https://askuma.hashnode.dev/mitigating-the-three-critical-llm-vulnerabilities-at-the-middleware-layer?utm_source=hashnode&utm_medium=feed)

[Jul 13](https://hashnode.com/posts/mitigating-the-three-critical-llm-vulnerabilities-at-the-middleware-layer/6a549903a93abc0fae91fdb5)

### [The Runtime Proxy Every Production AI Stack Needs](https://askuma.hashnode.dev/ai-guardrails-runtime-proxy?utm_source=hashnode&utm_medium=feed)

Why prompt engineering cannot secure enterprise LLM deployments—and what you should build instead.

5 min read · [askuma.hashnode.dev](https://askuma.hashnode.dev/ai-guardrails-runtime-proxy?utm_source=hashnode&utm_medium=feed)

[Jul 5](https://hashnode.com/posts/ai-guardrails-runtime-proxy/6a49f3264f0f6cecc8ce8026)