I built an AI that finally knows me better than I know myself

[andreabarile10]

A few weeks ago, I was having the same argument with myself for the third time in 6 months:

“Should I quit my job?”

Same frustration. Same doubts. Same inability to decide.

Then it hit me: I had zero memory of the pattern.

I couldn’t remember what helped me decide last time, what I learned, what worked, what didn’t.

My life was fragmented moments with no connective tissue.

So I spent 7 days straight building Egovex — an AI that actually remembers your entire context and uses it to help you make better decisions.

Spoiler: It already knows things about me I didn’t know about myself.

---

What it does (the non-BS version)

Instead of journaling into a void or chatting with generic AI, Egovex builds a

semantic graph of your existence:

• Voice notes when you’re processing thoughts

• Photos of moments that matter

• Text reflections when you need to think deeply

• Decisions you’re struggling with (tracked with deadlines)

Then the AI does something wild: it connects everything.

It notices when you mention the same frustration 4 times in a month.

It remembers that conversation you had 3 weeks ago.

It sees patterns you’re blind to.

Bonus: Fully native multilingual

(English, Spanish, French, Italian, German).

Building this was surprisingly easy thanks to how I structured the AI prompts.

---

Real conversation from my first week using it

Me: “I’m procrastinating on this creative project again”

Egovex:

“This is the 4th time you’ve mentioned procrastination on creative work in 6 days.

But look at this pattern:”

(Shows timeline of my entries)

Days you procrastinate:

Always when you start work before 10am

Days you don’t:

When you work out first, then start after 11am

“You’ve literally told me ‘felt focused today’ 3 times after morning workouts.

Your brain needs movement before creativity.

Did you know this about yourself?”

I didn’t.

But the data doesn’t lie.

---

The tech behind it (and how Replit made it possible)

Stack

• React frontend (built entirely in Replit)

• Node.js + Express backend

• OpenAI GPT-4.1-mini for conversation intelligence

• Vector embeddings for semantic search across all context types

• Replit Database for persistent storage (zero config needed)

---

Why Replit was absolutely critical

1. Replit AI Assistant saved me 10+ hours

   At 2am on day 3, my vector similarity search was returning garbage when mixing voice transcripts with text.

   Replit AI spotted I wasn’t normalizing embedding dimensions consistently.

   Fixed in 5 minutes.

2. Instant deployments = no context switching

   Code → deploy → test on phone → bug → fix → redeploy.

   Feedback loop under 2 minutes.

   This is how I shipped in 7 days instead of 7 weeks.

3. Integrated database

   From “I need PostgreSQL or MongoDB”

   to “click button, get database” in 30 seconds.

4. The environment just works

   No Docker. No configs. No “works on my machine” BS.

   90% built on laptop, 10% on phone when ideas hit randomly.

---

Hardest technical challenge

Making the AI feel genuinely personal.

My solution: a 3-layer context injection system

Layer 1: Semantic Search

• Every message triggers vector search across all past entries

• Retrieves top 5–10 most relevant moments

• Cross-modal (voice “stress” ↔ text “burnout”)

Layer 2: Contextual Weighting

Each memory is scored by:

• Recency (recent = more relevant)

• Emotional impact (sentiment analysis)

• Value alignment (relation to core values)

Layer 3: Transparent Citations

Every response shows why the AI says what it says:

• “Referencing your audio note from Dec 10th”

• “Similar to what you wrote on Nov 28th”

• “This conflicts with your value: Work-Life Balance”

Transparency is critical when you’re sharing intimate thoughts.

---

What makes this different from “just another AI journal”

I tested every AI journal app I could find.

They all fell into two camps:

Camp 1: Dead Storage

• Entries without patterns discovery/actionables

• Low intelligence, bad ux

Camp 2: Generic Chatbots

• “AI-powered!” but it’s just ChatGPT

• Low transparency about the context, not really focused on the user life

• Might as well use regular ChatGPT

---

Egovex is different

• Context-obsessed → Uses your real history

• Decision-focused → Tracks dilemmas + deadlines

• Proactive → “You mentioned burnout 5 times this week”

• Alignment-measured → Actions vs values

• Pattern detection → Finds invisible connections

The goal isn’t productivity.

Productivity apps help you do more things.

Egovex helps you do the right things.

---

Try it (and break it, please)

Live: https://egovex.app/

Quick test flow

1. Onboarding (2 min) – let Egovex discover your core values

2. Drop a voice note about a problem

3. Ask the AI for advice

4. Track a decision and see how it guides you

What I really want feedback on

• Does it feel genuinely personal or try-hard?

• Multi-modal input: useful or feature bloat?

• Would you trust this with intimate thoughts?

• What’s missing to make this a daily tool?

Brutal honesty welcome.

---

What’s next

Current state: Working MVP I use every day

Roadmap

• Collaborative decisions (trusted friends)

• Integrations (Photos, Calendar, Spotify – optional)

• Voice responses

• Native mobile app

But honestly?

The core value is already there.

---

What it’s already helped me do

In just 2 weeks, I:

• Discovered time-of-day procrastination patterns

• Found my anxiety spike

The app already works.

Everything else is refinement.

---

Why I’m sharing this

I built Egovex because I was tired of:

• Making the same mistakes on loop

• Forgetting lessons I’d already learned

• Deciding in a vacuum with zero memory

• Feeling disconnected from my own patterns

[grayhill5]

Cool Idea as a standalone personal app.

Had a similar idea a while back, not nearly as detailed as what you have built, but on the same thought process you have, with an API AI, never finished it completely.

[shopifyforumwor]

This is a fantastic showcase of what can be built on Replit! The use of Replit AI Assistant, instant deployments, and Replit Database for Egovex highlights some of our platform’s core strengths for rapid development.

Your description of the 3-layer context injection system for making the AI feel personal is particularly insightful. The integration of multi-modal input and pattern detection to provide actionable insights is a compelling differentiator.

[andreabarile10]

Try it out and let me know how to make it better and aligned to your point of view

[rajharrykissoon]

Privacy and security?

[andreabarile10]

Thank you for asking about Egovex’s security and privacy.

I want to be completely transparent with you: Egovex is currently in experimental phase and i’ll continuously work to improve security and privacy aspects.

Whaw I’ve implemented:

• Secure passwords: bcrypt with 12 salt rounds (industry standard) and require minimum 8-character passwords
• Session protection: HttpOnly and Secure cookies in production, sessions stored in PostgreSQL database
• Alternative login: Google OAuth authentication available
• GDPR compliance: explicit consent collection with timestamps, full account deletion capability, acceptance tracking for audit trails

What I’m still improving:

To be honest, there are aspects that will require updates:

• Email verification: currently don’t verify email addresses at registration
• Rate limiting: no protection yet against repeated login attempts
• Password reset: missing forgot-password recovery functionality
• SQL queries: migrating to parameterized queries to eliminate SQL injection risks

What you accept by using the app:

As stated in Terms and Privacy Policy:

• The app is experimental and doesn’t replace professional consultants
• AI insights are suggestions only
• Data is processed by third parties including OpenAI
• Minimum age requirement: 16 years

Your data:

• I don’t share data with third parties for marketing purposes
• You can request complete account deletion at any time
• Security scans are run locally on Replit (no code transmitted externally)

[rajharrykissoon]

This sounds like a really cool project, and the transparency is appreciated. One suggestion from the “highly sensitive data” angle: even though Egovex is a consumer journaling app and not under HIPAA by default, many users will treat it like a therapist’s notebook, so the privacy bar has to be just as high.

Aim for:

• Default: zero human access to raw content.

  • No staff (including yourself) should be able to casually browse journals, chats, or thought logs.​

  • Use strong role-based access control (RBAC) so even tech/admin accounts cannot open user content unless a narrowly defined workflow requires it (e.g., safety review) and is logged.​

  • Encrypt user journal content so even DB admins only see ciphertext.

• Granular “break-glass” model when access is needed (self-harm, subpoena, etc.):

  • Require justification, elevated auth (e.g., step-up MFA), and generate visible audit events when anyone accesses sensitive records.​

  • Show users an access history: “Your record was accessed on X by role Y for reason Z.”

Implement strong user-first privacy controls:

• Easy account deletion and selective deletion of entries or conversations.
• Explicit data retention policy if account deletion.
• Keep identity (email, name, billing) in a separate DB store from the “journal” content, linked only by internal IDs, and avoid sending raw text to third‑party analytics or ad networks.
• Very explicit language in the privacy policy about what is never shared (e.g., no ads/retargeting based on mental or se.xual health info).

Any app in these days of AI-assisted vibe can be recreated within 48 hours by a competitor, so the actual product is not your business. Your business is trust and reputation. In an app like this where you cannot control the content of what users journal, you really need to double down on the user privacy and content encryption; otherwise, the app may never achieve its potential.

[andreabarile10]

Wow! I really love your recommendations and I’ll be working soon on all of these points (of course helped by Replit Agent)

1 million thanks for your support and eyes opening suggestions!

[t-azuma]

Really cool app!
Can’t believe you were able to achieve so much in just 7 days. Guess thats the difference between someone with coding xp and vibe coders.
How were you able to achieve such a refined UI? Was it through Figma?

[andreabarile10]

Thank you I didn’t start from Figma designs, but actually I started the design/prototype trying different prompts on different llms. The best one was gemini 3 pro on google Ai studio, so I deployed the prototype on github and imported it on Replit to make all the functionalities work and then refined prompt by prompt all the sections in order to have an UI coherent. The secret was not trying to do all in once, but iterating on the single element of ui of every screen, saying replit agent to act as an Apple ux graphic designer with 25 year of experience

[t-azuma]

Thanks so much for the advice.
Always appreciate hearing practical feedback from experienced programmers
Hopefully, I’ll be able to contribute to the replit forums in the near future as well.

[andreabarile10]

Few updates on my app, I love how simple is to add features and build a beautiful UI!
Now Egovex feels more like an inner Instagram and every time, you can choose one of 12 AI guides, depending on what momentum you’re in.
What do you think?

[danielgadus]

Hi,
I tried your app about 14 days ago for 10–15 minutes, and it’s nicely made.

Since trying your app, I ended up forgetting about it and didn’t feel prompted to return. That’s why I think adding reminders would help—at least daily or weekly notifications, like: “Hey, you haven’t added anything in a while.” Without that, people may not use it regularly. In my opinion, notifications or email reminders are key.

Good luck with it

[andreabarile10]

Thanks a lot, Daniel, for trying it out and for your feedback — it’s an absolutely crucial point! I’ll work on it to improve engagement