DEV Community

Cover image for Have you done anything interesting with AI/ML?
Ben Halpern
Ben Halpern

Posted on

Have you done anything interesting with AI/ML?

Whether part of your job, school, or just spare time — have you made use of this branch of our field?

Discussion (35)

Collapse
dumboprogrammer profile image
Tawhid

Yes! made a tool to distinguish between ripe and unripe fruits.Unfortunately it's lost in the voids

Collapse
ben profile image
Ben Halpern Author

That's a pretty neat application. I'm sure this sort of thing is already becoming part of the grocery supply chain.

Collapse
dumboprogrammer profile image
Tawhid

yea nowadays, but back in the days this was pretty cool ngl

Collapse
vulcanwm profile image
Medea

woah thats really cool! how did u do it?

Collapse
citizen428 profile image
Michael Kohl • Edited on

I did a few interesting (to me at least) things, primarily around NLP. But they were all for clients so I don’t feel at liberty to talk about specifics. Ok, I gave a talk about one of them, so I guess that’s ok: I built a semantic similarity search, i.e. “you searched for A but B is a synonym/related concept so we also show you answers for that”.

For fun I built a classifier that would only show me positive news a couple of years ago.

Collapse
miguelmj profile image
MiguelMJ

Yes! To make a recommendation system for Stack Overflow unanswered questions

GitHub logo MiguelMJ / Answerable

Recommendation system for Stack Overflow unanswered questions

It even allows you to make your own recommendation model, so people who also have experience with ML/NLP can tune even finer if they want ;)

Collapse
bryce profile image
Bryce Dorn

I've been wanting to & over the weekend applied a deep learning model for facial recognition! Was pretty fun 😄

Collapse
guithomas profile image
Guilherme Thomas

I just finished my project on grain analysis (rice and lentil) to produce quality biodiesel. Used OpenCV so I could classify grains based on lenght-breadth ratio and color. First time doing a research like this but I had a lot of fun.

Collapse
ben profile image
Ben Halpern Author

Wow! Fascinating. What is next for you?

Collapse
guithomas profile image
Guilherme Thomas

I'll try to test in field. A local company is helping me with the grains to test and the next step is putting the code to work "in reality" haha

Collapse
theeasydev profile image
Uriel Bitton

Interviewer: What is your biggest strength?

Me: I am an expert in machine learning.

Interviewer: What’s 6 + 10?

Me: Zero.

Interviewer: Nowhere near. It’s 16.

Me: Ok, It’s 16.

Interviewer: What is 10 + 20?

Me: It’s 16.

Collapse
nombrekeff profile image
Keff

Most interesting thing I did was trying to make a music genre identifier using Tensorflow.

I got some results but got kinda bored, never been a big fan of ML. Don't like collecting and labeling data. I could've probably used some api or something to label the music, but couldn't be botherer if I'm honest.

Collapse
vulcanwm profile image
Medea

thats really cool! how did u make it?

Collapse
mburakerman profile image
Burak • Edited on

I've made AI Profile Picture Maker for AI people:
mburakerman.github.io/ai-profile-p...

Collapse
matthewvielkind profile image
Matthew Vielkind

I've been tinkering with a chatbot to help me discover / cook recipes. Right now I'm using mostly pre-trained NLP models for the document embeddings and semantic search, which work decent enough for now. I'm planning to fine tune the NLP models to improve search results and maybe add a voice integration with some QA models so I can ask "how much salt do I need?" and get a response instead of always looking back at the recipe.

An early prototype is available here. I mostly work with data on a day-to-day basis so this has also been a learning experience with front-end dev and application deployment!

Collapse
irreverentmike profile image
Mike Bifulco

Hi Ben! I'm just starting on a Livestreamed AI product - I'm building an app which uses GPT-3 to generate speeches for special occasions - weddings, graduations, retirements, things like that.

I posted about it here:

And I'll be livestreaming more work on it this week on my Twitch channel - twitch.tv/irreverentmike - the goal is to go from start-to-finish with a complete product idea in the span of a few weeks. I've got about 90min of work on it under my belt already... next steps are designing and implementing a usable UI, and hooking up API calls to it!

Collapse
adarshgoyal profile image
Adarsh Goyal • Edited on

Check out my blog what I have done with AI/ML at dev.to/adarshgoyal/my-journey-with...

also did this dev.to/adarshgoyal/eyeblink-detect...

Collapse
cchana profile image
Charanjit Chana

In terms of putting together models or training, no... but I was playing with the AI Art Machine earlier in the year.

On the waiting list for DALLE-2 so I can give that a try too!

Collapse
jrothlander profile image
jrothlander

I have a cross-language (trans) compiler I wrote years ago and I've working on adding AI/ML features to improve it. The most interesting adjustment I have made was to incorporate feature vectorizations on the parsing side and use that in code generation and optimization. I'm still discovering and researching how best to use AI/ML in these tools, and I don't work on it too often since it is a side project. But after only a few months I can already see advantages.

Think about it like this... feature vectorization of the syntax creates binary patterns that can be handled as image recognition problems. So you take the code and convert it to binary, as all compilers do, but instead of converting it to machine code, you vectorize it as features. Those features have patterns that can tell you how to handle code generation and that image recognition can address for optimization.

And yes, I am aware that compilers are not a good fit for traditional AI/ML per say. But we can use AI/ML in various parts of the process to make it work better and require less development efforts.

Collapse
matthewbdaly profile image
Matthew Daly

A few years back I wrote a letter classifier in PHP.

Before becoming a developer I worked for an insurer and they would scan any incoming correspondence and have staff classifying it, eg was it a complaint, request for information, request for surrender of their policy etc. I was interested in seeing how practical it would be to automate that. My solution involved taking a PDF file, converting it to a PNG, running it through Tesseract OCR to extract the text, and checking the text using ML to determine the intent.

My solution actually seemed to work reasonably well with my own mocked up examples, but if I was actually doing it for real there would be an extensive corpus of existing documents it could be trained on. Based on this I think it likely my old job won't be around by the end of the decade.

Collapse
kuwala_io profile image
kuwala

Many thing but there was one project which got not too good publications but was definitely a pop cultural thing that made a lot of fun .... Did NLP on Songtexts to detect emotions and topics and correlated it with google books mentions (cold back then ngram from Google a sort of Google Trends before the internet) to show the impact of political happenings on the topics and tonality of different bands

Collapse
maxfindel profile image
Max F. Findel

I developed an NLP Classifier for a client 6 years ago and I still maintain it. The tech stack has long gone stale and it could be in much better shape, but it still processes more than 100k customer surveys every month 🤷‍♂️

Collapse
j4ng5y profile image
Jordan Gregory

I would like to think so. Not to promote too much, but the company I'm a part of (plainsight.ai) is creating a nice platform for anyone to do Computer Vision ML/AI. I've used it a few times to help me more accurately predict when my friends and myself should duck and cover from severe weather :D.

Collapse
adriens profile image
adriens
Collapse
warwait profile image
Parker Waiters

I have messed around with PyTorch and some others libs, but have mostly been stuck in tutorial land.

Collapse
pratapkute profile image
Pratap kute

Not that much but going learn any suggestion?

Collapse
jzombie profile image
jzombie

Built a chat bot that usually always gives me bad advice: reshell.org/dr-reshell

Collapse
clpsplug profile image
C. Plug

Creating beatmaps (the stage data) for music games using ML. I'm looking into ways to improve the technique and that is where I started tinkering with ML.

Collapse
atulcodex profile image
Atul Prajapati

not yet :(

Collapse
saptakbhoumik profile image
SaptakBhoumik

I once made an app(1year ago) to detect lung disease from xray and ctscan. Link:- github.com/SaptakBhoumik/Palladium

Collapse
chetansingh63 profile image
Chetan Singh

Not yet! I am doing web development. But AI/ML is in our course curriculam in upcoming Summer Semester. So I will get introduced to AI/ML then.

Collapse
bluevalhalla profile image
Ed Miller

I work on an open source application to identify individual brown bears from photos and videos, BearID Project. The aim is to assist wildlife biologists in their research.

Collapse
onyeogulu profile image
Onyeogulu Tochukwu Rowland

Yes! currently working on detecting surgical instrument with yolov5 model

Collapse
zwadzinski profile image
wad

Trained a computer vision model for an autonomous drone boat that identified trash and identified objects to avoid!

Collapse
killerasus profile image
Bruno Baère

My master's dissertation was on the use of dynamic adaptive difficulty for shoot'em up games. The paper was published in 2013, already a bit old.

killerasus.itch.io/adaptive-shooter