finally coded up the attention part in “attention is all you need paper”

how was the experience ? awful

how many times did it fail ? near infinity,

but finally it work as the matplotlib graphs were showing up correctly.

tho i havn’t uploaded anything yet, but ill make all of this public soon ! 😭

but since i’m done with this, im thinking to build with somthing only via this library,

i’m thinking to build an sdk of kind. so others can just use it, also i wanna also train the model themselves without that much compute, ideas are invited.

i’m converting this from a scratch notebook to an actual library !

i’m implementing the encoder and decoder blocks and its soo much harder to actually implement when not following the guides 😭.

either way, i can’t wait to convert this full blown package. i’m thinking to implement the perceptron next.

Really sorry to all the people who voted on my project but couldn’t check out the demo.

a few users crashed the website.

this time i used auth tokens, reviewed rate limits and switched to render’s free tier (slow but generous !)

Changelog

• reverted to notifs
• fixed serverless 2
• fixed serverless
• made serverless
• fixes
• Delete pnpm-lock.yaml
• notifs
• updated client , and fixed bugs

previously, i implemented transformers from scratch with the help of Andrey karapathy and the atttention is all you need paper.

although i learnt a lot from this , i necessarily haven’t made anything useful.

hence i wanna make a python library that anyone can import and use to experiments around with LLMs.

this will mainly be a library of different architectures i wanna implement manually to actually get experience.

I spoke with the actual owner to understand what he wanted, after a while we finalized on a bold and dark design.

i thought it would be an easy task with ai (but the design it generated was utter slop.

i researched about good color match and made this design.

this was my second client project (which i’m doing for free to get some testimonials ),

more to GO !

I implemented qr code generators for kitchen staff would be able to generate QR codes that are specific to some tables

the users at those tables scan them.

there was a bug, and now the client side website also has call waiter button that now does send a notification and a banner to the backend.

the waiters can now know which table they should go to.

Also, i fixed a bug where the additional notes field doesn’t seem to send anything to the kitchen staff.

Changelog

• feat: QR code generator, waiter call notifications, order notes in kitchen

squished some mega gnarly bugs,

first off, i fixed the chat box! the text kept trying to run away past the bottom border like it was escaping prison… bad text, you stay in your box!!

anyways, my brain is fried. (plz don’t eat me)

Changelog

• did it dirty
• fix: loading animation and silent failed responses
• fix: animations delay and slow response
• fixed no reply bugs
• using gemini api key now
• Update README.md
• pages
• all systems working
• persitent bugs
• visuals bug fix
• added genarate api, but more like fixed it
• added agents
• api added haiku
• removing replit deps
• Saved progress at the end of the loop
• Enhance the learning experience with an improved animation system and an interactive mentor
• Transitioned from Plan to Build mode
• Saved progress at the end of the loop
  ….and a lot more

i built a brain for it, i hooked up the AI mentor chat thingy so now you can literally talk to it while you code! LMAO, atleast you don’t need to switch tabs to talks with chatgpt ig!

it was super glitchy at first…

it kept spitting out huge chunks of text that looked like alien math.

added a sparkly icon because we like shiny things over here.

the bot is actually pretty smart, definitely smarter than me.

wazzup gang!! building coding learning thingy!

this project mostly just me throwing spaghetti at the wall to see what sticks.

but at one point the background was white ?!
my eyes are still burning ngl

anyways, you can type words in it now, which is a big win for humanity.

worked on the kitchen and admin dashboard, since i already planned it all out in the backend, it wasn’t a big problem now.

but it does work now:

order something here: https://qr-cart-six.vercel.app/menu/1 (please use phone view)
and you would be able to see it here: https://qr-kitchen.vercel.app/ (use the correct pin)

dont actually use this project as a production deploy, maybe dm me and ill put out a verison for you !

Having the client and admin apps in separate repos was becoming a massive headache for shared types,

so I just merged them into a proper monorepo using git subtrees (atleast learnt something new).

It feels so much cleaner having everything under one roof.

I also reorganized the Firebase rules into their own folders.

It’s finally starting to look like a professional project instead of two separate experiments.

Used web sockets for sync between the customer’s phone and the kitchen tablet.

Seeing a “New Order” pop up on the KDS the exact second I tapped checkout on my phone was such a pain to implement 😭!

Used Firestore listeners to keep things easy.

Still fighting some Tailwind layout issues on smaller screens, but the core flow is solid.

im thinking to build a toggle so the chef can kill items if they are out of stock. No more disappointed customers ordering sold-out pasta!

Changelog

• chore: update firebase.json with new firestore paths
• chore: update root .gitignore with build and IDE patterns
• chore: organize firestore rules and indexes into firebase/ folder
• chore: relocate project documentation to docs/ directory
• build: add root package.json for monorepo orchestration
• build: configure pnpm workspaces for client and admin apps
• docs: add comprehensive project README with deployment links
• feat: setup public directory for QR code storage
• feat: initialize shared directory for common utilities
• Remove submodules to prepare for subtree merge
• new dashboard

Utilizing Gemini’s extravagant free tier to scan privacy policies.

Changelog

• feat: add gemini support to backend
• updated to gemini

Building websites that Dominate Search.

basically offering good ui and seo !

how i do it: ai genrated designs are bad, they look trash and drown people’s trust in the product, i instead sourced the components from 21.dev and react bits, who provide free componets for stunning websites.

while also maintaining good performance and data structure on the site to appeal to the great mother google.

Changelog

• push
• contact update
• price update
• faq fix
• works
• features is a bit broken
• still working
• works !
• first commit
• Initial commit from Create Next App

Hercule design doc:

A browser extension that makes AI read boring privacy policies. This system analyzes privacy policies and presents them through an intuitive interface with actionable insights.

a few other features:

• it has a scoring for each website which determines your security (0-100)
• Highlights concerning privacy practices
• Personalized recommendations with one-click email templates
• Automatically finds privacy policies on any website
• Results are cached for instant repeat visits
• Automatic key rotation for rate limit resilience