An Industrial-Grade Showcase for the "New Year, New You" Google AI Portfolio Challenge

Deployed to Google Cloud Run AI Engine: Gemini 3 Flash Secured by Cloudflare

Production URL: https://ayushpandey.me
Infrastructure Status: Shielded via Cloudflare Edge (DDoS Mitigation)

Most portfolios function as static galleries. This platform is an Engineering Audit.

I have architected this portfolio to expose the "How" and the "Why" behind my development ecosystem. Instead of a surface-level list, I provide a deep-dive into my flagship SaaS, VersionCV. Using Gemini 3 Flash to generate structured React Flow pipeline, I created a dynamic map of distributed service dependencies, allowing evaluators to audit the system design of my project in an interactive way.

React Flow Architecture Figure 1: Interactive System Design view of projects generated via Gemini 3 Flash and React Flow.

The portfolio itself is optimized for high-performance delivery while maintaining professional-grade security:

• Shielded Origin Architecture: The portfolio's Google Cloud Run instance is protected by a Cloudflare Proxy. This setup masks the origin IP, mitigates DDoS risks , and prevents "Denial of Wallet" attacks by filtering malicious traffic before it impacts the compute budget.
• Design Decision FAQs: I have integrated a technical FAQ for project to explain the intuition behind my architectural choices.

VersionCV leverages Gemini 3 Flash for its core resume-to-job-description optimization engine.

• Hallucination Control: I implemented a strict XML Tagging schema for prompts. This algorithmic approach enforces structured data outputs and ensures high-fidelity entity extraction.
• Performance Benchmarking: R&D indicated that while various LLMs offer quality outputs, Gemini 3 Flash provided the lowest latency and highest cost-economy for high-token NLP tasks.

I utilized Gemini to bridge the gap between abstract architecture and visual implementation within this portfolio. It assisted in generating the node/edge logic for the interactive React Flow pipelines that map out the internal services of the projects I build.

This portfolio chronicles a real-world performance refactor on Google Cloud Run:

1. The Bottleneck: Initial monolithic deployments suffered from a ~5 second cold-start latency when the service went idle.
2. The AI-Powered Audit: Using Gemini to analyze Docker layer efficiency, I identified that unoptimized dependency overhead was bloating the image size.
3. The Solution (3-Stage CI/CD): I refactored the deployment into a Multi-Stage Docker Build:
   • Stage 1 (Deps): Strict isolation of package.json to maximize layer caching.
   • Stage 2 (Builder): High-performance compilation environment.
   • Stage 3 (Runner): A minimalist node:20-slim production image.
4. The Outcome: Drastic reduction in image pull times and optimized Scale-to-Zero configuration, resulting in significantly decreased cold-start times and zero idle costs.

• AI/ML: Google AI Studio (Gemini 3 Flash), Gemini CLI
• Infrastructure: Google Cloud Run (Primary Compute), Cloudflare (WAF/DNS)
• Frontend: Next.js (App Router), Tailwind CSS, Framer Motion, React Flow
• Runtimes: Node.js