Skip to content
View Ayush-cloud06's full-sized avatar
🎯
Focusing
🎯
Focusing

Block or report Ayush-cloud06

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
Ayush-cloud06/README.md

Ayush Kumar Yadav

Cloud Security & GRC Engineer
AWS · Terraform · Policy-as-Code · ISO/IEC 27001

Asansol, India · Open to remote
official.ayush.066@gmail.com · ayushcloud.dev · github.com/Ayush-cloud06 · LinkedIn


I turn compliance requirements into Terraform, OPA/Rego policies and CI gates that leave audit evidence behind. I'm an ISO/IEC 27001:2022 Lead Implementer (TÜV SÜD) and an AWS Solutions Architect – Associate, currently preparing for AWS Security – Specialty.

Featured project

Ayka Secure Technologies GmbH: compliance-as-code for a simulated EU SaaS company

Note

Ayka is a fictional 24-person SaaS company in Stuttgart, used as a case study. Each part below is labelled with its real status: runs in CI, designed (Terraform written and validated, never deployed), or documented (draft GRC records).

flowchart LR
    PR[Pull request] --> P[terraform plan<br/>mock credentials]
    P --> S[Checkov · OPA · tfsec]
    S --> E[Evaluator<br/>38 ISO 27001 controls]
    E -->|any HIGH| F[Fail]
    E -->|any MEDIUM| A[Human approval]
    E -->|LOW only| OK[Pass]
    E --> EV[(Checksummed<br/>evidence bundle)]
Loading

DevSecOps pipeline · runs in CI

  • Policy gate on every pull request. Checkov and OPA/conftest read the Terraform plan JSON, tfsec reads the source, and a Python evaluator maps every finding to one of 38 controls referenced to ISO/IEC 27001:2022 Annex A.
  • Three-way, fail-closed decision. HIGH fails the build; MEDIUM pauses at a GitHub environment until a human approves; LOW passes. Missing, malformed or empty scanner output counts as a failure, and an unmapped finding defaults to MEDIUM so a person always reviews it.
  • Proof that the gate catches problems. A deliberately insecure workload (public S3, open SSH, IMDSv1, no encryption, open NACL) has to fail with specific controls reported by specific tools, otherwise CI goes red.
  • Audit evidence from every run. Each run produces a bundle: the plan, raw scanner JSON, a Markdown report, and a manifest that records the commit, tool versions and SHA-256 hashes of the mapping, the evaluator and every Rego file. The apply step re-checks those hashes before it runs.
  • Exceptions with accountability. Every waived finding needs a reason, an owner and an expiry date, and it still shows up in the report.
  • A hardened pipeline. Tools are pinned by checksum and actions by commit SHA. The token is read-only, CI holds no cloud credentials, gate logic is covered by CODEOWNERS, and main requires 5 passing checks.
  • Tested end to end. CI runs 48 Python tests and 37 Rego tests. I found and fixed a wrapper bug that silently dropped every tfsec finding, then added a test so it can't come back.

Identity & access · designed

  • Joiner-to-access chain in Terraform. An HR dataset feeds Microsoft Entra ID users and department/tier groups. SCIM provisions them into AWS IAM Identity Center, and tiered permission sets lead into cross-account IAM roles.
  • Tiered privilege. Three permission sets (Platform-Admin, Tier1-Ops, Tier2-Workload) with 8-hour sessions and a 1-hour session for high-privilege work. Entra Conditional Access policies require MFA for Tier 0 and block legacy authentication.
  • ABAC and escalation control. Department isolation compares aws:PrincipalTag/Department with aws:ResourceTag/Department, with mandatory tagging and permission boundaries to block privilege escalation.
  • AWS Organizations guardrails. Security, Infrastructure and Workloads OUs, with SCPs that stop CloudTrail from being disabled, deny root-user activity and restrict Regions. Break-glass roles come with a written procedure.

GRC · documented

  • ISO/IEC 27001:2022 ISMS. Scope, information security policy, risk methodology and criteria, a register of 13 risks scored 5×5 (inherent, current and target), plus a treatment plan and an acceptance log. One control (A.8.24, S3 encryption) is traced end to end: risk → requirement → Terraform → test → CI evidence.
  • SOC 2. System description, Trust Services Criteria scoping (Security, Availability and Confidentiality in scope; Processing Integrity and Privacy out, with reasons), control matrix and 7 policies. The readiness assessment concludes the company is not ready for a Type I and explains why.
  • NIS2. Applicability assessment under Germany's NIS2UmsuCG/BSIG: with 24 staff, Ayka is below the threshold but still inherits NIS2 duties through customers' supply-chain requirements. Includes an Art. 21 gap assessment and an Art. 23 incident-reporting procedure.
  • GDPR. Art. 30 record of processing, Art. 28 processor register, and Art. 32 measures for identity and segregation of duties. I found that the region SCP (ap-south-1) contradicts the EU data-residency target and logged it as a risk.

Other projects

Cloud Platform Control Plane · Terraform module

  • Hardens a bare AWS account in one apply:
    • MFA-gated, role-only IAM.
    • A permission boundary that denies all IAM actions and denies disabling CloudTrail, GuardDuty or Config.
    • Multi-region CloudTrail with KMS, log validation and optional S3 Object Lock.
    • 15 alarms for CIS AWS Foundations v3.0.0 controls 4.1–4.15.
  • Every control maps to CIS, ISO 27001 and NIST CSF with an evidence command an auditor can run. CI runs 27 terraform test cases against a mocked provider, plus Checkov → SARIF, TFLint and gitleaks.
Earlier projects
Project What it does Stack
Compliance-Gated Deployment Pipeline Checkov before apply, human-approved apply over OIDC, Prowler + OPA after, evidence in versioned S3 GitHub Actions · OIDC · Prowler · OPA
AWS Automated Remediation Guardrails Config and GuardDuty findings trigger Lambda fixes: block public S3, revoke open SSH, delete root keys EventBridge · Lambda · SNS
Cloud Policy Engine Rego rules for Terraform plans and live AWS state OPA · Rego
Cloud Security Compliance Automation IAM access-key and S3 ACL audits with CSV/JSON evidence Python · boto3

Technical skills

AWS IAM, Identity Center, Organizations & SCPs, CloudTrail, CloudWatch, Config, GuardDuty, Security Hub, KMS, S3, EventBridge, Lambda, SNS, Firehose
Identity Microsoft Entra ID, SCIM, Conditional Access, RBAC & ABAC, permission boundaries, break-glass access
IaC & CI/CD Terraform (modules, feature flags, terraform test, TFLint), GitHub Actions, OIDC federation, environments & rulesets
Policy & scanning OPA/Rego (conftest), Checkov, tfsec, Prowler, gitleaks
GRC ISO/IEC 27001:2022, SOC 2 (TSC), NIS2 / BSIG, GDPR, CIS AWS Foundations, NIST CSF 2.0
Languages Python (boto3, pytest), Bash, HCL, Rego · Docker (basic)

Certifications

Certification Issuer Date
ISO/IEC 27001:2022 Lead Implementer TÜV SÜD · Cert. No. IN/62172/645925 Sep 2026
AWS Certified Solutions Architect – Associate Amazon Web Services Jul 2026 · valid to Jul 2029

Pinned Loading

  1. Ayka-Secure-Technologies-GmbH Ayka-Secure-Technologies-GmbH Public

    Compliance-as-code gate for Terraform plans: Checkov, tfsec and OPA findings mapped to ISO 27001 Annex A controls, a fail-closed pass/approval/fail decision, and checksummed evidence. Plan analysis…

    Python

  2. aws-automated-remediation-guardrails aws-automated-remediation-guardrails Public

    The AWS Automated Guardrails System ensures that security rules are continuously enforced through pipelines, detection systems, and automatic remediation.

    HCL

  3. aws-security-engineering-core aws-security-engineering-core Public

    A modular infrastructure framework that models how detection, response, identity, compliance, and protection controls are architected and deployed in AWS.

    HCL

  4. Compliance-Gated-Deployment-Pipeline Compliance-Gated-Deployment-Pipeline Public

    GitHub Actions pipeline that enforces security and compliance policies before Terraform deployments using Checkov, Prowler, and OPA (Rego).

    Python

  5. cloud-platform-control-plane cloud-platform-control-plane Public

    The Cloud Platform Control Plane is a modular, feature-flag-driven security framework that transforms raw AWS accounts into hardened, audit-ready environments. It establishes an immutable security …

    HCL

  6. Cloud-Policy-Engine Cloud-Policy-Engine Public

    OPA-based policy engine for cloud security, compliance enforcement, and preventive governance

    Open Policy Agent