Building a Secure "Personal Gemini Journal" — Gen AI Academy APAC C3

Nikita Parjane — Sun, 06 Sep 2026 19:55:43 +0000

Submission for the Google Cloud x Hack2Skill Ideathon Challenge: Build a Secure "Personal Gemini Journal" — Gen AI Academy APAC Edition, Cohort 3.

The Challenge

Most AI-generated apps look great in a demo and fall apart in production — hardcoded keys, no auth boundaries, shared databases with zero isolation. The challenge: configure Google AI Studio to think like a security engineer before writing a single line of code, then ship a real, production-grade application on top of that foundation.

Phase 1 — Configuring AI Studio as a Security-First Foundation

Before generating any code, I set up Google AI Studio with custom System Instructions establishing a standing "constitution" for the build: threat-model every feature before implementing it, never hardcode secrets, enforce per-user database isolation by default, and apply secure coding standards throughout.

Phase 2 — What I Built

A full-stack Personal Gemini Journal: React (Vite) frontend, Node.js/Express backend.

Real Firebase Authentication — email/password sign-in via the Firebase client SDK. Every backend API call carries the resulting ID token, verified server-side with the Firebase Admin SDK before any data is touched.
Multi-turn AI journaling — real conversations with the Gemini API, with cascading model fallback for resilience.
Isolated Firestore storage — every user's data lives at /users/{uid}, keyed strictly by the server-verified UID — never a client-supplied field. Matching firestore.rules enforce the same boundary.
Secure key management — the Gemini API key is designed to be retrieved via Google Cloud Secret Manager at runtime, never hardcoded.

Phase 3 — Feature Enhancements Beyond the Base Spec

Multi-tab conversations — each chat tab is an independent thread, so a user can hold several parallel reflections without losing context.
Smart RAG semantic search — query past entries by meaning, not keyword matching.
Autonomous goal-extraction agent — synthesizes unstructured journal logs into milestones, roadblocks, and a prioritized action checklist.
Emotional trend analytics — aggregated positivity scoring with an on-demand Gemini-generated deep-dive.

Architecture

Dockerized with a multi-stage build (React frontend compiled and served by the same Express container as the API), Cloud Run–ready, health-check-safe on Cloud Run's injected $PORT.

Repo: https://github.com/Nikita-Parjane27/personal-gemini-journal-fullstack
Video walkthrough: https://youtu.be/ECT_7hqBh3w

AccelerateAIwithCloudRun #GenAIAcademy #GoogleCloud #Hack2Skill