shubhamkakkar

The Problem

Standard practices say no non-root process gets to talk to the Internet on a port less than 1024. How, then, could I get Node talking on port 80 on EC2? (I wanted it to go as fast as possible and use the smallest possible share of my teeny tiny little micro-instance's resources, so proxying through nginx or Apache seemed suboptimal.)

The temptingly easy but ultimately wrong solution:

Alter the port the script talks to from 8000 to 80:

}).listen(80);

shubhamkakkar

// make sure you have not installed react-native vector icons by now

make a file named react-native.config.js, content should be following
```
module.exports = {
    projects: {
      ios: {},
      android: {},
    },
    assets: ['./assets/fonts'],

shubhamkakkar

version: 2
aliases:
  - &restore-yarn-cache
    name: Restore cached root node_modules
    key: yarn-cache-{{ checksum "yarn.lock" }}
  - &save-yarn-cache
      name: Save node_modules to cache
      key: yarn-cache-{{ checksum "yarn.lock" }}
      paths:
        - node_modules  

shubhamkakkar

wversion: 2.1

node_image: &node_image
  docker:
    - image: node:12

android_image: &android_image
  docker:
    - image: reactnativecommunity/react-native-android

shubhamkakkar

file=path/to/websiteVersion.json

increment_patch_version='increment_patch_version'
increment_fix_version='increment_fix_version'
increment_feature_version='increment_feature_version'

commandToRun=${1:-increment_patch_version}

featureVersion=`cat ${file} \
  | grep featureVersion \

shubhamkakkar

How to use nginx as a reverse-proxy with letsencrypt

Your infrastructure

generated via plantuml

Requirements

shubhamkakkar

🛡️ AI Guardrails: Zero to Hero

A Complete Guide for the Modern AI Practitioner

"Guardrails aren't barriers to AI progress — they are the infrastructure that makes safe, sustainable AI innovation possible." — IBM Think

---

📋 Table of Contents

shubhamkakkar

Threat	Description
Prompt Injections & Jailbreaks	Adversarial inputs that manipulate AI behavior to produce restricted or unsafe outputs
Sensitive Information Exposure	Outputs that include PII, proprietary data or sensitive information such as healthcare records
Misinformation & Harmful Content	AI-generated outputs that spread false information, toxic language or biased perspectives
Unpredictable Model Behavior	LLMs that generate unexpected or unsafe outputs without proper safeguards
Open Source Vulnerabilities	Risks that arise when open source AI models and APIs lack sufficient guardrails for safe use
Unfiltered User Input	Instructions from end users that push AI systems beyond intended limits, leading to unsafe or harmful outputs

shubhamkakkar

Check Type	What It Does
Prompt Injection Detection	Catches "ignore previous instructions" patterns
PII Scrubbing	Strips emails, SSNs, phone numbers from input
Topic/Intent Classification	Ensures query is within allowed scope
Token Length Validation	Prevents context overflow attacks
Jailbreak Pattern Matching	Regex/classifier on known bypass attempts

shubhamkakkar

Check Type	Description
Streaming Token Monitoring	Flag/stop generation if toxic content emerges mid-stream
Agent Action Validation	Validate each tool call before execution in agentic flows
Confidence Scoring	Pause if model confidence drops below threshold
Loop/Recursion Detection	Catch infinite agent loops before they escalate
Budget/Resource Limits	Kill long-running LLM chains to prevent runaway costs