This is a submission for the GitHub Finish-Up-A-Thon Challenge
What I Built
Havendew is an AI-powered journaling system built around one idea: your journal should learn who you're becoming,
not just store what you wrote.
I've journaled for as long as I can remember, thus this project is very personal to me. It started the way it does for most people — writing about my day, one entry at a time on paper.
But somewhere along the way, it stopped being that.
I started watching self-improvement videos and something shifted. I started writing differently. Not just recording my day but interrogating it. What did I learn today? What patterns am I repeating? What do I actually want? Then gratitude sections. Then goal reviews. Then affirmations written out by hand every single day. Then manifestation -scripting my future in present tense like it had already happened. Then identity tracking. Then shadow work. Then weekly reviews where I'd read
everything back and ask: what was I avoiding? What emotion was actually running things this week?
Every few months I'd tear the system apart and rebuild it. Keep what worked. Remove what felt performative.
Eventually I had something that wasn't really a journal anymore. It was a personal operating system. One I'd spent years building prompt by prompt, section by section. And for years I had one thought sitting quietly in the back of my mind: what if this wasn't trapped in notebooks?
A while ago I built the first version of that idea. It was basic — you could write an entry and save it. That was it. It worked technically but it had none of the soul of the actual system. Life got busy. The project sat there unfinished.
The GitHub Finish-Up-A-Thon gave me the reason to go back.
Not to polish the basic version. To finally build the
real one.
The Comeback Story
Before:
A single text input. A save button. An entries list
that said "No previous entries yet."
That was the entire app.
After:
The design went from plain HTML to a full premium system — warm cream gradient canvas, liquid glass bento cards, forest green Soul Report that feels like a completely different room, editorial Cormorant Garamond
typography, polaroid photo memories, affirmation cards with soft gradients.
The daily entry became a 5-section scrapbook workspace:
- Daily Gratitude with quick chips and photo memories
- Today's Chapter — the main free-writing canvas with rich text, stickers, and auto-save drafts
- What I Learned — lesson extraction
- Daily Affirmations — rendered as beautiful gradient cards
- Extra Prompts drawer — shadow work, pattern analysis, identity proof, emotional analysis hidden behind a collapsible toggle
The AI layer appears only after you save — never before. It generates: Guidance From Today (warm, specific to your actual words), Growth Pattern, Tomorrow's Invitation, Gentle Affirmation, and a Letter From Your Future Self. All from Google Gemini reading what you actually
wrote, not a generic template.
The Soul Report lives on a full forest green canvas (#24352C). It reads like a letter. Four sections with colored left borders: 🌿 Growth Pattern, 🍃 Hidden Root, 🌾 Gentle Truth, 🌱 New Growth. Three personalized prompts for next week. A closing affirmation in gold Cormorant serif. It feels like receiving a letter from the wisest
version of yourself.
The Character Arc groups entries by week, calls Gemini to name each era in 3-4 words (Quiet Progress, Builder Era, Finding My Rhythm), and displays them as a vertical timeline with proof tags and dominant mood.
The Visual Trajectory shows 4 charts from your actual data: mood flow line chart, journaling rhythm bars, moments of growth area chart, dominant moods.
The Identity tab is a scrapbook profile with your initial, desired identity badge, streak counter, 4 operating modes (Builder, Study, Reset, Career), goals tracker, and JSON export.
The most important decision in the entire build:
Writing first. AI second. Always.
Every AI journal I've tried interrupts the writing with prompts and analysis before you've written one honest sentence. Havendew waits. You write your gratitude, your chapter, your lessons, your affirmations. You close the day. Only then does the AI speak.
That one decision made Havendew feel less like software and more like the journaling system I actually wanted to exist.
Demo
GitHub: https://github.com/aashitanegii/journalAI
Live Demo: https://havendew-881393616978.us-central1.run.app/
Demo YT Video https://youtu.be/ecBR1ldITy8
My Experience with GitHub Copilot
Copilot didn't just help me write code faster. It helped me actually finish this time.
1. The schema architecture & Firebase migration
My entry model needed nested morning/evening blocks, identity proof arrays, pattern intelligence tags, anti-waste metrics, and an auto-calculated efficiency score. I described the structure in plain English. Copilot helped me seamlessly migrate the entire architecture from a local database to a scalable Google Cloud Firestore structure, writing the queries and the efficiency score calculation logic on the first attempt.
2. The AI prompt engineering
To return reliable structured JSON — shadow prompt, pattern note, morning/evening gap detection, affirmation — required very precise system prompting. The tone had to be right: not a therapist, not a chatbot, but the wisest version of the user talking back. Copilot helped iterate the system prompt for gemini-1.5-flash until the output was consistent enough to render directly in the UI.
3. The bento dashboard
Four distinct card types — glass, dark, warm amber, sage — in a responsive grid, each with its own visual language. Copilot generated all four variants and the grid composition cleanly. I spent my time on the
content inside the cards.
4. The weekly report aggregation
Compiling 7 days of nested entry objects into a single weighted prompt for the Soul Report was the most complex logic in the codebase. Copilot wrote the array-mapping function and suggested weighting shadow work entries more heavily than micro-wins when building the summary — which actually improved the AI output quality.
5. The character arc era naming
The entries list groups entries by week and asks to name each era in 3-4 words. Copilot generated the grouping logic and the API call structure in under 10 minutes. It's the feature people will remember most
and it almost didn't make the build.
6. The Visual Trajectory charts
Four stacked Recharts graphs with real data streaming directly from Firestore. Copilot generated the data-mapping functions that convert raw entry arrays into the shape Recharts expects — mood strings mapped to numeric values, entries grouped by week for the consistency bars, and cumulative proof counts for the area chart.
Without Copilot the gap between "basic text box app" and "full AI identity journaling system" would have been too wide to close in 6 days.
Tech Stack
- Frontend: React 18 + Vite + TailwindCSS + Recharts
- Backend: Node.js + Express.js REST API
- Database: Firebase Firestore
- Auth: JWT + bcryptjs
- AI: Google Gemini API (gemini-1.5-flash)
- Deployment: Google Cloud Run
- Design: Cormorant Garamond + Inter, custom glass card system, forest green Soul Report palette
I've kept journals in notebooks, apps, notes apps,
Google Docs, voice memos — everything. None of them
did what I actually needed.
Building Havendew felt less like shipping a project
and more like finally giving the system a home it
deserved.
It took years of notebooks to figure out how to
journal well.
It took 6 days and GitHub Copilot to build it.
Top comments (0)