Engineer · Founder · Open Source

Ashish
Khanagwal

Building the enforcement layer between AI and production systems.

B.Tech AI & Data Science @ MAIT Delhi · Founder of Croofx · CNCF Contributor. Focused on what happens after AI writes the code validation, safety, and production reliability.

See Croofx → Get in touch @TheAshrex ↗
scroll to explore

The story

I build things that enforce discipline on AI.

It started in 2nd year at MAIT, contributing to a human and weapon detection research project that ended up published in IRE Journals and got into Kavach 2023, India's national public safety hackathon run by the government. I learned early that the hardest engineering problems aren't always the flashiest ones.

Since then I've contributed to Meshery (a CNCF sandbox project), built AI pipelines at VertexPlus processing vehicle telemetry in real-time, maintained open-source repositories through Hacktoberfest, and founded Code for Community a developer community that grew to 400+ members without any marketing budget, just genuine care for helping other engineers grow.

Today I'm building Croofx a deterministic enforcement engine that sits between AI-generated code and your CI pipeline. Because AI is getting very good at writing code. But no one's built the brakes yet.

400+
Community members led
97★
Open source repo stars
8.51
CGPA · B.Tech AI & DS
2023
IRE published

Current build

Croofx | The AI Execution Control Layer

Founder · Solo · Building in public
AI code is getting merged. Nobody's actually checking it.
Every team using AI to write code has the same invisible problem: the AI generates patches probabilistically, but your CI pipeline, your production systems, your deployment contracts, they demand determinism. Right now, developers are manually acting as that enforcement layer. Reviewing AI output line by line. Hoping nothing breaks. Croofx automates that layer. It sits between your AI tool and your codebase, validates every patch before it touches production, and blocks the merge if it fails with a hard exit code, no exceptions.
AI says: "I think this patch is correct."
Croofx says: "Validation failed. Merge blocked."
$ croofx validate --ci
→ scans repo → validates patch → FAIL → exit(1) → pipeline fails → merge blocked

No configuration needed. No GitHub OAuth required. Works offline except for the AI call. No AI in the validation path.
CLI · Python CI/CD enforcement Patch validator Model-agnostic Merge-blocking
Zero AI in validation path v0.0.1 in progress

How it’s built

Core Enforcement Engine
Deterministic repository analysis built for repeatable results. Croofx scans and models the structure of your codebase so AI-generated changes can be evaluated in a stable, predictable environment before they ever touch your workflows.
v0.0.1 → v1.0.0 in progress
Foundation layer
Structured AI Contract Layer
AI output is normalized into strict structured formats before it reaches the validator. Croofx turns loose model responses into concrete change contracts that can be safely analyzed and verified.
v0.0.2
Schema enforcement
Patch Simulation & Validator
Croofx simulates AI-generated patches against an in-memory copy of the repo, checking structure, syntax, and dependencies for hidden breakage. Only deterministic, validator-approved patches move forward there is zero AI inside the enforcement path.
The nucleus
v0.9.0 pre-release
Roadmap: SaaS + Governance
The long-term layer that gives teams policy control and visibility over how AI touches their code. Model routing, usage policies, and audit trails all flow through a single governance plane.
v2.0.0
Long-term vision
Join the waitlist →

Work

Where I've built.

May 2025 — Present
Layer5 · Meshery
MLOps Engineer
  • Contributing to Meshery, a CNCF sandbox project service mesh adapters, cloud-native tooling, multi-cloud compatibility.
  • Engineered AI-driven observability systems, metrics processing pipelines, and real-time model performance analytics.
  • Participating in MeshMate mentorship, CI workflows, and community-driven engineering improvements.
KubernetesCNCFDistributed SystemsMLOps
Oct 2024 — Mar 2025
VertexPlus Technologies
AI Engineer Intern
  • Built an AI conversational assistant using Phidata, LangChain, and Ollama to generate actionable insights from vehicle telemetry.
  • Python-based ETL pipelines that reduced data processing latency and improved prediction accuracy by 25%.
  • Deployed backend microservices integrating AI models with production APIs for continuous inference and monitoring.
LangChainOllamaPhidataETLPython
Aug 2023 — Sep 2024
Accessible For All
Software / AI Engineer
  • Optimized data ingestion and preprocessing pipelines in Python, NumPy, Pandas reduced processing time by 30%.
  • Built automated CI/CD pipelines via Jenkins, GitHub Actions, and GitLab CI, cutting deployment cycles by 40%.
  • Engineered RESTful APIs serving ML-based accessibility recommendations with 20% improved real-time responsiveness.
PythonCI/CDJenkinsDockerREST APIs
2022
Hacktoberfest
Repository Maintainer
  • Reviewed and merged 80+ pull requests, enforcing code quality and structured branching strategies.
  • Automated workflows using GitHub Actions, improving CI reliability and contributor efficiency.

Projects

Things I've shipped.

Open Source Practice & Resources
Community-driven resource hub covering DevOps, ML, Web3, backend, and frontend. 391 commits. Grew organically during Hacktoberfest no marketing, just genuine utility.
★ 97 stars · 91 forks GitHub
Penguin Classifier ML Web App
End-to-end ML application in Python and Streamlit. Random Forest model with 98% accuracy. Single and batch inference modes, confidence score visualization, CSV upload support.
Scikit-learn · Streamlit · Pandas
Finance Data Pipeline (Simulated)
Microservice-based pipeline using Java, Golang, and PostgreSQL. Full containerization via Docker Compose with health checks. Automated builds with Makefile and CI workflows.
Java · Golang · Docker · PostgreSQL
NBA EDA Dashboard
Streamlit dashboard analyzing NBA player stats. Automated web scraping with BeautifulSoup, interactive correlation heatmaps, comparative performance charts.
Python · Streamlit · Seaborn · BeautifulSoup

Research

Published work.

IRE Journals · Vol. 6, Issue 11 · 2023
Smart Street Light System Using IoT & Computer Vision
Built the core human and weapon detection algorithm using Python and OpenCV. Integrated AWS IoT Core for device connectivity and telemetry. Deployed Jenkins and GitHub Actions CI/CD pipelines. The project got into the Kavach 2023, India's national government-run public safety hackathon.
PythonOpenCVAWS IoT Core Computer VisionCI/CDKavach 2023
Read the paper ↗

Community

Beyond the keyboard.

⚙️
Code for Community
Founder
Founded and scaled a 400+ member community focused on AI/ML and Data Science. Organized workshops with Eduhub Community, created open-source repositories, and mentored developers building real-world projects.
📡
IEEE Week of Code
Project Administrator
Guided teams building AI/ML and web development projects in a competitive engineering environment. Mentored contributors on Git workflows, Python, and data pipeline design.
🌐
Layer5 · Meshery
CNCF Open Source Contributor
Active contributor to a CNCF sandbox project. Working on service mesh adapters, cloud-native tooling, and CI workflows alongside a global engineering community.
🔨
Building in Public
@iashishkhangwal on X
Sharing the real story of building Croofx the decisions, the bugs, the thesis. Raw, technical, no hype. Threads, takes, and the occasional 💀 when AI agents break prod.

Skills

The toolkit.

AI / ML
Python TensorFlow PyTorch Scikit-learn LangChain Ollama OpenCV Pandas NumPy
DevOps / CI-CD
Docker Kubernetes GitHub Actions Jenkins GitLab CI Terraform Ansible Helm
Backend / Systems
Golang Node.js Java REST APIs PostgreSQL MongoDB Bash Linux
Cloud / Infra
AWS (EC2, S3, Lambda) GCP DigitalOcean Prometheus Grafana Kafka Apache Airflow

Contact

Let's talk.

Whether you're a founder, recruiter, open source contributor, or someone thinking hard about AI in production. I'm always up for a real conversation. No cold decks. No spam. Just builder talk.

Interested in Croofx?
If you're a team that ships AI-generated code to production and wants deterministic enforcement before it reaches CI, let's talk. Early access, feedback sessions, or just a technical deep-dive.
Talk about Croofx →
Open to opportunities
MLOps, DevOps, AI Engineering roles where I can build real systems. Especially interesting if you're a startup doing something ambitious.
Send a note →