Devansh Malhotra

While this redesign took a little longer than I originally expected, the result was worth the hard work. I have completely replaced the original AI-generated interface with a custom, hand-crafted design that, hopefully, improves on the aesthetics and overall user experience.

Since the entire frontend codebase is now my original work, I am removing the AI declaration. I chose to keep most of the HTML code as it was already coded by me, and the emojis help in easier navigation, but the styling and design language are now 100% original.

🎉 MailMate is now 100% human-coded and designed by me. 🎉

NEXT STEPS:

• Custom Labels: Let users create their own categories (e.g. “Education”).
• Kill the 10-email limit: Let the AI ingest the whole inbox at once.
• Fix the To-Do bug: “No reminders found” error even when reminders exist.
• Clickable Charts: Make the pie charts interactive so clicking a slice filters the inbox.
• Daily Digest: Have MailMate auto-send a daily to-do list to the user’s email every night.
• Search by Date: Add a proper way to filter/search for emails from specific days.
• Mobile Cleanup: Fix those weird layout issues on phones.
• AI Noise Control: Make the classification smarter so it stops creating reminders for every trivial thing.
• Keyboard Shortcuts: Rapid-fire shortcuts (e.g., ‘E’ to archive, ‘T’ for to-do) for power users.
• AI Chat Search: A bar where you can just ask “When’s my flight?” and the AI finds the answer.
• Dark & Light Modes

This project actually started as my Internship Capstone Project. You can check out the original “v1” demo here -> https://www.bettermindlabs.org/devansh-malhotra
(and yeah, please ignore how I look in that video. They did me dirty with everything, I promise I look better IRL!).

The original build was a great proof of concept, but the Gemini API kept hitting 429 rate-limit errors and breaking my email classification. Over the last ~3 hours, I’ve been grinding to turn this into a legit, production-ready app. Here’s what I have been cooking:

AI Migration: Ditched Gemini for Groq’s LLaMA 3.3 70B.
Prompt Batching: Smartened up the processing logic to group multiple emails into single API calls. Requests dropped by ~80%, and processing speed skyrocketed.

Multi-User OAuth: Rebuilt the entire auth flow. Instead of a single-user desktop script, MailMate is now a proper web app with multi-user support via web-based Gmail OAuth (with automatic token refreshing).

Classification Fixes: Worked on the classification logic to handle empty bodies and weird formatting that used to break the sentiment/spam detection.

What I’ll be working on now:
Design Refresh: The UI looks really bad, even though it was AI-generated. I will spend time stripping out the messy gradients and chaotic animations for a “humane” design system.

🎉 SentinelNet Final Devlog🎉 :

1. Fixed Interface Demo 404s
2. Split & Repository Launch:
   • v2.0 Repository: Cut the core engine into 2 main pieces. All “Remote Access” (the Tunnel, Service, and Remote Desktop API) now lives in its own dedicated home: SentinelNet_v2.0.
   • v1 Cleanup: Purged all RAT-like code and 150MB+ of binary bloat from the main repository’s history, reducing it to a high-speed, basic network scanner.

All future updates for SentinelNet_v2.0 will be published in its dedicated project devlog.

I just wrapped up the Meds Module. I’ve moved beyond simple tracking to create a truly intelligent clinical assistant. Now, MediSafe can take those complex lab results and instantly suggest a personalized medication and supplement plan using AI.

I focused heavily on the user experience here, which makes clinical data feel approachable. You can now seamlessly bridge the gap between AI recommendations and doctor-prescribed reality by editing, removing or manually adding meds to your profile.

P.S. I know using emojis for the app’s navigation icons might make people think it’s AI-generated, but I think they make the UI way easier to navigate. It’s a deliberate design decision to keep the interface intuitive, and it is not AI-generated!

I didn’t realize that adding devlogs every two hours was mandatory, so I’m logging the full session here. You can check my GitHub commit history to see the progression of the work and confirm that everything was developed manually.
During this time, I worked on SentinelNet, building and refining the core networkscanner module for device discovery and port/service scanning. I debugged module imports, improved the project structure, and cleaned up the architecture to better separate scanning, analysis and reporting components. I also tested functionality across different scenarios to ensure stable and predictable behavior.

I’m working on my first project! This is so exciting. I can’t wait to share more updates as I build.