AI for Personal Use That’s Actually Production Ready

Introduction

Background and Role

Thank you so much, Moenstone, and first of all, I would like to thank Carl, who invited me for the first time, and then connect me with Amna and Grete, and yeah, basically, I am actually a bioinformatician, if you don't know about bioinformatician, it's basically doing some solving biological problem with, like, data analysis, but yeah, the office is in Tartu, in iCosogen cell factory.

And I did my PhD actually in Thaltech. So it's just next door.

And been working in iCosogen as a bioinformatician slash full -stack developer because you'll see about what I've done.

But I'll start with this.

Foundations of Bioinformatics

So who knows about these sequences and stuff like that? DNA, RNA, protein sequences?

No? No one?

Oh, yeah, there are some. some.

The Central Dogma Refresher

Okay, but if you don't know about the sequences, this is like a central dogma, like the thing that you always talk about the molecular biology and biotechnology from DNA, RNA, and then amino acids.

From Sequences to Tools

And these kind of sequences that I've been like doing, analyzing, and basically making the tools like user -friendly web applications and doing these sort of alignments.

And surprise,

AI and Secure Internal Tools

Surprise, surprise, because of AI, I have to establish our own internal tools, because as you know, in healthcare, you can't really share your data, so data security is like a forefront thing. So we currently use the Ollama, if you know about this.

Yeah, that's really cool. And then, yeah, it's DocRise, and we're using the OpenWeb UI.

Stack Choices and Constraints

but again another surprise I won't talk much about this instead I will share

Weekend Projects and Practical Builds

like I said I'll share more towards like what I've been doing including like making my own stuff during like weekends because you want to spend your time really nice and then in this AI era you only have like some split windows to kind of do your hobbies and doing your stuff so I'd like to start with this so

Starting From Problems and Requirements

basically it's always start with problems requirements or it's just part

Simple Sites: Resumes and Blogs

of the job right you need to do something and then the simplest stuff is just landing pages like resume personal blogs and stuff like that and basically you'll send this to your clients or even like yourself to put everything in there

Tooling: VS Code, Copilot, GitHub Actions

And with this, I've been using like Yes Code, GitHub, Copilot, and then GitHub Actions, and it's already done. It's actually just written using these things.

And if you scan the resume, you can check everything there. So I don't host it anywhere else, just GitHub page for this one.

let me check this one yes probably this one yeah so wow okay do you see it now

Showcase: Personal Site and Integrations

so yeah as you know it is pipe coded and uh because i i started with mostly python during my phd so i didn't really know about javascript and stuff like that and basically it's just like

Like I want to put my stuff including like research because I have a research background and then some other kind of stuff like blogs. I write a lot in Medium at least the past months and some of them require more specific treatments.

APIs for Medium and Scholar

like you need to call an API for instance for medium because you can't really just embed it and then also for Google Scholar it's a bit tricky because you need another API just to scrape it and then basically put it there and that

Why Copilot Helps Keep It Cheap

was like super easy and this is the part where I want to show you like yes the copilot and why i suggested this because everything that's here because this is my personal project

i want to put it as uh as cheap as possible i would say and i want to to see whether uh it's it works or not and and so far it is really good if if you have already like a template you have a good abstraction like the things that you want to kind of uh put in your like resume

or website especially in the static website so there is not much but you can again check or probably ask if something specific about this like for instance this having this medium and then how to deploy it like it's super simple and you can just ask like it is quite interesting that

New IDEs and Favorite Models

a few days ago, Google also released their own, like, IDE, right, anti -gravity, but it works pretty much the same. It's just that, yeah, it is an extension in a copilot, and

you can, like, have lots of different models to kind of write. My favorite is actually Grokode.

Not sure if anyone also a big fan of Grok, no? Or is this me?

Sorry. Yeah, okay,

Okay so that's the first like how I use it and then actually publish and I've never had any problem it's just that sometimes you just need to kind of set when the actions this is actually the first time also I learned like Crony can actually run in GitHub action and it's really nice you just need to set the tokens and stuff like that and yeah with along with AI you can kind of have a good

Building a More Complex App: Sweatlogger

assistance okay that's the first and then the second sometimes like you need

more complex applications like this is the part where to me it's a bit funny

because I never I I always have a problem going to the gym so is there

anyone also have the same thing okay so i'm not alone okay but during this vibe coding yeah when

The Problem: Forgetting Gym Reps

i tried like doing more codes like with more coding uh in terms of not part of my job uh i tried to find some problems which is like when i'm doing like some exercise in the gym sometimes i

forgot my reps and this is super annoying you know like having like three straps of squats and then like i always think like did i do twice or it already thrice and these things like keep happening

and i thought probably i should fix this by building my own like logger right so you can

Implementation: Hosting and Auth

check this uh qr code this is actually again it's just build vs code github and copilot but i didn't use uh github action because it is meant for a static so you need something more dynamic so i use Versal for hosting the web application and basically we can check hopefully it works

looks like the file coded but yeah it is log i mean because i didn't want to be bothered with the like login and authentication so i just used google it's it's quite simple quite straightforward

and this is like the thing that i try to kind of manage right so you can have a template late uh like yeah i stuck in fourth month fourth month so it's it's uh i couldn't really put more weights uh because like i got sick like two weeks ago as well can get the history and then all the summary i even can share this with the the people that already set up my uh work working plans so

Mobile Experience and Features

So this is quite nice, but the more interesting part is that I already, or like let's say, this is already adjusted to the mobile web application. So as you see, you can check, can get the first template running, and yeah, like for instance, the squad, yeah, I'm quite weak. So probably it's just, oh no, this is just the warmup.

I can show you my core squad, yeah. Oh, I need to close this one first. See, there's a bug.

Yeah, like this. And then you can do the timer.

And when I wanted to kind of, okay, so I don't want to work out with this. I can just call the AI.

AI-Assisted Workout Plans

So do I need to do, probably I want to push day or pull day or both, right? And then just write like, let's say I only have 30 minutes. Let's work out.

Hopefully the API works. And based on this, it's actually calling Google. I'm not sure whether it's a Gemini 2 .5 Pro still. still and then it returns a JSON and then basically from the JSON I put it.

Nope. Yeah. OK, so this is embarrassing, but not let's say one of these.

Again, I just spend like an hour or two hours for two or three weekends for this. But hopefully it works. don't worry the first demo was the same okay but yes so basically it should give me

uh it should give me the sets everything like this yes okay so probably the box if i put four so three sets two sets but i did 15 reps and stuff like that and basically like you have except up and then you will have these things and it is ready and then once you're done and then you can

record it and then uh well basically you have your own analytics and stuff so yeah it's a bit geeky but i like it and this motivates me to go to the gym this is the part like the best part of inventing this sweatlogger so yes this is just an example and I want

Reflections on Learning in the AI Era

well do I still have time okay so yes but now for the last part actually I

From Problem Discovery to Solution Narrowing

want to emphasize like before the AI era like we already or we need to learn a lot of things first right we need to learn even about the problems the requirements and then demands and and and even like the requests we just need to know and then sometimes we branch

on with peers we get loads of lots of solutions and then we start narrowing it down and getting the exact solutions that we want to have but now with ai we have questions ai and then it's sorted it instead of like oversimplifications that actually some people really like it but sometimes

Limits of AI-Only Coding

when we are getting used to it it's just that we will take a lot of things like for instance coding with AI only with AI you will kind of put a lot of technical steps in it and if you don't really know exactly like the structure infrastructure about the framework that you are using then this

is this is really annoying at the end and I said like if if if I can give an

Architecture Matters: Scaling and Costs

example for this like the the sweat lover for instance I used serverless database so it is in progress right for for productions then it's just need to be scaled like even like for hosting I need like to think about the load

Balancer if many people are using it right and then first of all actually have I need to subscribe and then it scales and and this is the the small little things that we need to actually know from

the beginning if we don't really know then we will have a lot of problem at the end like yeah I stole

like slides from Vitaly he said engineering of Bactum no I actually wrote the message that I

Beyond UI: Systems and Agents

I'll use like this slide that actually AI is just beyond UI, UX or landing page creations or even web applications. And it is more than like building something behind the things that people are talking about, even like I just talked.

and and as you see I really like this slide because like yeah mostly people are stuck with the quick UI UX because it is pretty quick and it is super easy to show people that what you can build what is your idea and what you want to kind of deliver and you can have feedbacks pretty quick

You can always have digital twins. You can have agentic stuff.

and now like people start building different agentics even like I have sometimes then I built built my own logger so that I don't want to like subscribe a lot of things. And from there you can see a lot of potential as well.

Conclusion

So and yeah, actually, that's my last slide.

So thank you very much.

Finished reading?