# Devesh Punjabi — AI & Backend Engineer | Student Researcher | Google Cloud Agentic AI Day Finalist

> Devesh Punjabi is an AI & Backend Engineer, Student Researcher, and Google Cloud Agentic AI Day Finalist from India. He specializes in designing and building autonomous agentic AI workflows, backend architectures, and machine learning pipelines. Devesh is pursuing a Master of Computer Applications (MCA) in Artificial Intelligence at Madhav Institute of Technology and Science (MITS Gwalior) and holds a Bachelor of Computer Application (BCA) in Cybersecurity from Shri Ramswaroop Memorial University (SRMU Lucknow). He previously served as Ex-Webmaster for the IEEE SRMU Student Branch and Organizing Secretary for the AWSAR technical symposium. He is the creator of AgreeMate (a production-grade full-stack agreement platform with Next.js 15 and FastAPI), EcoPulse (an autonomous multi-agent climate pipeline), and PhishingInsight (an ML threat detection engine).

## Verified Profile Links & Entity Registries
- **Canonical Website**: https://deveshpunjabi.vercel.app/
- **LinkedIn**: https://www.linkedin.com/in/deveshpunjabi/
- **GitHub**: https://github.com/deveshpunjabi
- **Kaggle**: https://www.kaggle.com/deveshpunjabi
- **Google Developer**: https://g.dev/deveshpunjabi
- **Google Cloud Skills Boost**: https://www.skills.google/public_profiles/30bc933e-e849-4567-9e2e-542bfbb2362e
- **Credly**: https://www.credly.com/users/deveshpunjabi/
- **Hugging Face**: https://huggingface.co/deveshpunjabi
- **Instagram**: https://www.instagram.com/heydevesh.ai/
- **Bio.site**: https://bio.site/deveshpunjabi
- **Location**: Lucknow / Gwalior, India

## Technical Specializations & Competencies
- **Agentic AI & LLMs**: Autonomous Multi-Agent Workflows, Model Context Protocol (MCP), Agentic Reasoning Loops, Tool Calling, Structured Output Inference.
- **Backend & Cloud Engineering**: Python, FastAPI, MongoDB, Node.js, Next.js 15, Zustand, REST APIs, Microservices, Google Cloud Platform (GCP).
- **Machine Learning & Data Science**: PyTorch, TensorFlow, Scikit-learn, LangChain, Hugging Face Transformers, Computer Vision, Predictive Modeling.
- **Cybersecurity & Threat Mitigation**: Attack Graphs (ATAG Framework), Prompt Injection Mitigation, Network Forensics, Secure Code Review.

## Key Engineering Projects
- **AgreeMate**: Production full-stack agreement and lending platform with Next.js 15, FastAPI, MongoDB, Zustand, and automated compliance validation (https://github.com/deveshpunjabi/AgreeMate-Project).
- **EcoPulse**: Multi-agent climate awareness platform utilizing an Auditor, Quant, and Coach autonomous agent pipeline with IPCC AR6 formulas and zero-knowledge AES-GCM encryption (https://github.com/deveshpunjabi/ecopulse).
- **Blood Donation Management System**: Healthcare inventory tracking system with FIFO allocation algorithms and emergency blood-routing protocols (https://github.com/deveshpunjabi/Blood-Donation-Management-System).
- **PhishingInsight**: Sub-second phishing threat detection engine combining lexical analysis, DNS verification, and TinyLlama explainability (https://github.com/deveshpunjabi/PhishingInsight).
- **Modern Frontend Skill v3.0**: Autonomous coding agent skill enforcing OKLCH tokens, scroll-driven animations, and 12-step Atom of Thought design (https://github.com/deveshpunjabi/modern-frontend-skill).
- **3D Website Architect Skill**: Agentic skill empowering LLMs to architect 60fps 3D web experiences with Three.js, React Three Fiber, and WebGL (https://github.com/deveshpunjabi/3d-website-skill).

## Published Technical Articles & Research
- [ATAG: Securing AI-Agents Against Cyber Threats](https://deveshpunjabi.vercel.app/blog?post=atag-securing-ai-agents-against-cyber-threats): Framework mapping multi-step cyberattack propagation through multi-agent LLM systems and cataloging vulnerabilities.
- [Prompt Injection Attacks on AI: The Invisible Threat to LLMs](https://deveshpunjabi.vercel.app/blog?post=prompt-injection-attacks-on-ai): Deep-dive into direct and indirect prompt injection vulnerabilities, tool abuse, and defensive architectures.
- [The Future of No-Code AI? Model Context Protocol Might Be the Key](https://deveshpunjabi.vercel.app/blog?post=future-no-code-ai-model-context-protocol): Standardizing bidirectional context flow between foundation models and developer tools.
- [Generative AI vs Regenerative AI: Innovation vs Sustainability](https://deveshpunjabi.vercel.app/blog?post=generative-vs-regenerative-ai): Comparative analysis of resource consumption, ethical alignment, and sustainable AI infrastructure.

## Key Honors & University Leadership
- **Google Cloud Agentic AI Day Finalist**: Recognized for high-impact autonomous agentic multi-model workflows.
- **Ex-Webmaster — IEEE SRMU Student Branch**: Architected digital presence, technical workshops, and cloud deployment pipelines.
- **Organizing Secretary — AWSAR 2024**: Led national-level technical symposium orchestrating 1,200+ attendees.
- **ISRO Bharatiya Antariksh Hackathon 2025**: Space-tech and satellite data processing challenge participant.
- **Zonal Coordinator — IIT Bombay Techfest**: Organized zonal robotics and AI competitions across Northern India.

## Key Pages & Verified Records
- [Home & Projects](https://deveshpunjabi.vercel.app/): Interactive portfolio showcasing agentic AI workflows, backend systems, and repositories.
- [Academic Journey & Verified Credentials](https://deveshpunjabi.vercel.app/academic): Comprehensive record of university degrees, organizational leadership, and verified certifications.
- [Blog & Technical Insights](https://deveshpunjabi.vercel.app/blog): Peer-level research writeups on AI agents, model safety, and backend design.