PRASAD TILLOO

Posted on Feb 2

I Built a Portfolio That Thinks Like an Architect (Using Google Gemini + Cloud Run)

#devchallenge #googleaichallenge #portfolio #gemini

New Year, New You Portfolio Challenge Submission

This is a submission for the New Year, New You Portfolio Challenge Presented by Google AI

An experience-driven AI portfolio that matches client challenges against 15+ years of real delivery patterns — not generic ChatGPT responses.

👋 About Me

I’m Prasad Tilloo — an independent Enterprise Architect and Transformation Consultant based in Germany who's spent 15+ years helping enterprises navigate cloud migrations, AI adoption, and compliance-heavy transformations. I've worked with everyone from healthcare giants to climate tech startups.

But here's the thing: every client asks the same question - "Have you done something like this before?"

That question inspired me to build something different. Not another generic AI chatbot, but a portfolio that actually thinks like an architect.

Portfolio

🌐 Cloud Run App link:

🎥 Quick Demo Video (2 minutes):

👉 Try it yourself at https://prasadtilloo.com/tools/project-similarity

Key Screens

Homepage

Similarity Matcher

How I Built It

The Core Idea: Experience-Driven AI

Most AI portfolios just slap ChatGPT onto a website. I wanted something smarter.

My system analyzes your project description against structured signals from real projects:

Industry patterns
Technical constraints
Anti-patterns I've observed
Decision frameworks that worked
Retrospective lessons

Google Gemini 1.5 Pro handles the reasoning, but it's constrained by real project metadata - no hallucinated architectures.

Tech Stack & Google AI Integration

Frontend: React + TypeScript + Tailwind

Backend: Node.js on Google Cloud Run

AI: Google Gemini 1.5 Pro

Data: Google Sheets (lightweight CRM)

Email: SendGrid for lead capture

AI-Assisted Development Workflow

I used Google Gemini + Antigravity in a "vibe coding" approach:

Gemini for architectural reasoning and refactoring suggestions
Antigravity for rapid UI iteration
Human decisions for UX structure, domain modeling, and production hardening

Example prompt I used:

"Build experience-driven matching using project metadata, not embeddings. Score industry, constraints, anti-patterns, and decision frameworks. Return top 3 with confidence."

This accelerated development while keeping architectural decisions manual.

Production Architecture

🏗️ System Overview:

Container-level architecture showing Cloud Run hosting both frontend and API, Gemini-powered similarity engine, Google Sheets CRM, SendGrid email delivery, and Namecheap DNS with managed SSL.

Key Components:

Frontend: React + TypeScript + Vite SPA
Backend: Node.js API on Google Cloud Run
AI Engine: Google Gemini 1.5 Pro for similarity matching
Data Storage: Google Sheets for CRM + Static JSON for projects
Email: SendGrid for transactional delivery
Infrastructure: Custom domain + SSL via Cloud Run

Data Flow:

User describes project challenge via React interface
Cloud Run API processes request and queries project database
Gemini AI analyzes similarity patterns against 15+ years of experience
System generates personalized insights and recommendations
SendGrid delivers results via email after lead capture
Google Sheets stores lead information for follow-up

Browser (React + Vite)
↓
Cloud Run (Node API)  
↓
Gemini 1.5 Pro
↓
Project Similarity Engine
↓
Google Sheets (Leads + Tool Requests)
↓
SendGrid (Email Delivery)

Why Cloud Run? Zero infrastructure ops, automatic HTTPS, simple CI/CD. Perfect for a consulting business.

Custom Domain Setup: Namecheap DNS → Google Cloud Run with automatic SSL certificates.

What I'm Most Proud Of

1. It's Actually Useful

This isn't a demo. It's my real business system. Clients use it to understand if their project matches my experience before booking calls.

2. Experience-Driven AI Differentiation

Instead of generic responses, visitors get:

"Here's the closest project I've done like yours — what worked, what failed, and what I'd do differently today."

3. Production-Grade Implementation

GDPR-safe lead capture
Rate limiting on AI endpoints
Feature flags for staged rollout
Email gating before AI results
Proper error handling and fallbacks

4. Strategic Focus

I intentionally enabled only one AI feature for this submission. Why? To showcase architectural thinking over feature dumping. The Project Similarity Matcher demonstrates real business value, not AI novelty.