gemma-4-good-hackathon

[astral saddle]

So it's 100% necessary to make our code public in order to win?

[tacit eagle]

Hey quick question. Could an education project be made for stduents rather than educators?

[tacit eagle]

I think so

[cunning shore]

Yo Max want to team up

[deep hemlock]

Looking for team mates

[real zodiac]

Hey hii

I'm abhi

[deep hemlock]

Interested can dm

[fallen maple]

Yes

[lone crow]

Hi folks, we are looking for 2 team members to join us in the gemma-4-good-hackathon .DM if interested

[pine fox]

hi, I look for team members in competition gemma-4-good-hackathon , preferably with kaggle best practices experiences (as i am noob in kaggle) . I have already finetuned the model . Here the project draft : Project Name: ZenithHealth
Mission: Decentralizing Medical Intelligence for the Last Mile.
DM me to see how we can collaborate.

[lone crow]

The project idea is indeed interesting...

Can U share the code with the coding section in the challenge!

[tepid oar]

Wait i thought it was for students only!

[tacit eagle]

I thought educators looking at the rules...

[wraith oasis]

Hay, it's Amit!
Looking for a team for Inter College Hackathon in India..

[pine fox]

did any one else completed the fine tuning ? i did it with HF pipeline on my hardware. had to overcome a few issue but the resulting model performs great. finally i tested it also in ollama

[spiral bolt]

Hey I'm looking to join a team for an upcoming hackathon. I bring strong skills in Python and Rust, with a focus on AI/ML, building models, and developing MCPs. If you're putting together a team and need someone who can move fast on both the modeling and integration side, let’s connect

[misty iris]

Hi I’m looking for teammates and looking to compete in the Health and Sciences track

[trim tulip]

What are you looking to build?

[misty iris]

I’m looking to build either a simulator for biological experiments or something with health data analysis (i.e. Taking in healthcare data and outputting actionable insights that are personalized)

[lime bane]

Hi team,

Quick clarification on prize eligibility — the rules state that projects are eligible to win both a Main Track Prize and a Special Technology Prize. Does this mean a single entry can only win one prize from each track? Or is it possible to win multiple prizes within the same track (e.g., both the Ollama and Cactus prizes from the Special Technology Track, or both the Future of Education and Digital Equity prizes from the Impact Track)?

Thanks in advance!

[digital robin]

For the llama.cpp Special Technology Track: would a submission centered on a measured efficiency improvement to llama.cpp inference with Gemma 4 (with proper benchmarks and reproducible setup) score better on its own, or does it need to be wrapped inside an application use case (e.g. "Education with llama.cpp") to be judged favorably?

[regal hare]

Just finished working on a project and am quite impressed on how gemma 4 2b parameter model can be quantized to 3gb of memory size but still capable of giving remarkably amazing reasoning ,wow

[regal hare]

Meaning you can confortably run the model locally on a 4gb ram phone giving room for the os to handle its operations

[novel veldt]

Hello I wanna is there anyone uses claude with libraries like
Caveman ,claude-mem,obsidian
Or with olama i wanna know how to utilise with maximum efficiency

[unkempt quest]

anyone exploring AEC routes for gemma 4?

[karmic mural]

I5 11th gen 16gigs ram with 500gb HDD, i'm running ollama with openclaw and i need to user a model for ollama, which one can i use for more faster results and maximum efficiency?

[grave hollow]

You still working on this? I was working on something similar

[cedar zinc]

What tracks are y’all exploring?

[ornate token]

RAG and NLP. what about you?

[rigid coral]

Hey, so I had this project idea for a separate hackathon that I competed in about two weeks prior to the start of this competition. I was planning on working on it afterwards but never got the chance until I came across this hacathon. Would it be against the rules if I repurposed the project to use Gemma 4 and other Google tools? I technically started the project before this hackathon but would need to do a complete rebuild from scratch anyway.

[unique badge]

Hey guys,

I'm a second year B.Tech student and a beginner for this hackathon, I would like to join hands with a team for this hackathon, is there anyone would like to join me into their team.

If anybody interested kindly DM me.

Thank You!!!

[digital robin]

any suggestions to this? would be very much appreciated xD

[drifting crest]

Since Codex runs on the cloud, would I still be able to use it as a tool or would that not be allowed?

[radiant shadow]

Hello, I am developing a tool for the hackathon focused on the local use of llm

LLS (LLM-Local-Stack)
LLS (LLM-Local-Stack) is a sleek, free, open-source no-GPU AI framework that turns ordinary CPU + RAM into a private, vendor-free engine for running local LLMs anywhere.

This project is a local LLM framework that combines the best ideas of Amazon Bedrock and LM Studio for local deployment. The repository supports GGUF model inference through Python proxy layers, making local models much easier to integrate into real applications.

It also includes a TUI management tool for simple and effective model control, helping users manage models and services more easily.

The framework is highly flexible and designed with zero vendor lock-in. It can work with observability frameworks and integrate with a wide range of AI tooling, including LangChain, LangGraph, MEMO, Coggne, and RML, among others.

https://gitlab.com/jacobofp2001/llm-local-stack-gemma-model.git

[real zodiac]

Good bro

[unique badge]

Hey guys, hope all are doing great.

I'm looking for a team and couldn't figure out how to find a team on Kaggle.

[near lotus]

Hey everyone! I am Shashank, ex-Amazon, IIT Guwahati alumni, and a vibe coder who has shipped multiple AI products. Currently exploring the intersection of AI and biology, drug discovery, and personalized medicine.

Building Project Rosie for Gemma 4 Good hackathon: https://www.shashankpadala.com/blogs/project-rosie-cancer-vaccine . A man used ChatGPT and AlphaFold to design a cancer vaccine for his dying dog and it worked. I am trying to build the infrastructure to make that repeatable.

Looking for collaborators with background in computational biology, bioinformatics, vet oncology, mRNA synthesis, or someone who can animate the science the way cartoon show "Ozzy and Drix" explained the human body as a kid. Wanna use that as inspiration for creating project demo video. If any of the above describes you, send me a friend request here and DM me.

Happy to be here, excited to see what everyone is building!

[hard storm]

Hey @everyone I am looking for the team for this hackathon

[carmine falcon]

Hi

Do the LiteRT versions of Gemma4 eg https://huggingface.co/litert-community/gemma-4-E4B-it-litert-lm/blob/main/README.md support multimodality?

I'm getting errors using gemma-4-E4B-it-web.task in JS for multimodal tasks, but gemma-4-E4B-it.litertlm works in Python.

Sample code would be great, but otherwise an idea if this is my problem or not would help.

[carmine falcon]

There error I'm getting is

INVALID_ARGUMENT: Calculator::Open() for node "LlmVisionInferenceCalculator" failed:
Image models could not be created

[smoky arrow]

Hello, I am looking fo members for gemma-4-good-hackathon.

[rose vector]

I'm a solo participant in the Gemma 4 Good Hackathon. My project runs fully offline on local Ollama + Gemma — it's a voice-intake tool that processes audio locally on a MacBook Air with no cloud dependencies. This is core to the project's design (data sovereignty for humanitarian use cases).
Two quick questions on submission requirements:

For the Live Demo requirement — since the app requires local Ollama + a 14GB Gemma model and can't be hosted as a public URL, is a Kaggle Notebook that demonstrates the pipeline acceptable? Or would you prefer uploaded project files?
For the Code Repository — I'm planning to attach a public GitHub repo. Any issue with that vs. a Kaggle Notebook for the code portion?

Want to make sure I'm not accidentally disqualifying myself on a technicality. Happy to adjust format if needed.

[native sail]

👋 Hi everyone! I'm Steve - working on FarmWise AI 🌾

A climate-smart agronomist for 500M smallholder farmers, powered by Google Gemma 4.

What it does:

• Diagnoses crop diseases from leaf photos (94% confidence)
• Works 100% OFFLINE on $50 Android phones
• Speaks Vietnamese, Hindi, Spanish, French... 20+ languages
  Why it matters: 500M farmers produce 70% of the world's food but face 20-40% crop losses due to lack of expertise.

• Check it out: https://www.kaggle.com/code/solokop/farmwise-ai-agricultural-assistant

Would love feedback from the community! 🙏

[rose vector]

FYI @slow lark @restive raft

[unkempt quest]

1 week left lets gooooooo

[restive raft]

All questions should be asked in the forums, hosts do not monitor discord.

[rose vector]

I tried to - https://www.kaggle.com/competitions/gemma-4-good-hackathon/discussion/698239

I just dont want to be DQ'd for not understanding

[tall saffron]

Hello I am looking for someone who can help me with building tool . Please send dm

[native sail]

I Just published our writeup for FarmWise AI — offline agronomist for 500M farmers powered by Gemma 4.

20 demos, real inference, multilingual advisory. Check it out! https://www.kaggle.com/code/solokop/farmwise-ai-agricultural-assistant

[weak fern]

hi guys!

is submission under all three tracks is allowed?

[unkempt phoenix]

yeah

[fiery laurel]

Anyone struggling in Consistency and want to learn together.
DM me.

[native sail]

for using an AI model?

[native sail]

Yes, I think so!

[weak fern]

thank you !

one of the stuff actually confirmed that it's okay

[waxen pier]

hey guys

[pine dagger]

Hola!

[pastel dagger]

hello

[wind topaz]

Hi admins, Couple of team members dropped since they needed to focus on exams. How can I remove them? Please provide guidance

[crimson brook]

I am looking for teamates for google gemma competition

[lean totem]

Notebook and vision looks nice!

[native sail]

Please +1 Vote to support me

[unkempt quest]

gahhhhhhhh theres less than 7 hourssssss gahhhhhhh

[whole sluice]

damn we couldnt submit due to wifi issues, i dont suppose submission could be extended a bit? (prob not but doesn't hurt to ask)

[pliant rampart]

"Your Writeup should not exceed 1,500 words."; Mine is 6,036 Words long.....

i had an oversight and just realized it 5 mins ago

Literally jumped from my bed

im so cooked 😭 🙏🏻

oh well, regardless, im very proud of this project and one of the best that ive done so far, i even put my words in closing notes, i think that's why i got into 6k words long but yea, win or not, im standing proud with the results

https://www.kaggle.com/competitions/gemma-4-good-hackathon/writeups/Swara

[fallen maple]

Does anyone know when the result will be declared?

[pliant rampart]

uhhhh

[rich sand]

Over 1600 submissions, it will take a few months

[pliant rampart]

the last gemma3n hackathon took 4 months to be announced

and its 4 judges with 600 submissions

so expect it to be a long judging phases

[pliant rampart]

maybe ill start a discussion thread abt it later, and hopefully judges can estimate the expected time, i wouldn't mind waiting even for half years if communication is clear hahah

[coral granite]

But you deserve thanks from the community for substance. Well done!
I didn‘t know the writeups are already public and had assumed otherwise. That‘s so useful to know when posting on social media. 👍

[buoyant anchor]

Hello everyone. I had a question about working on a project post-competition. I submitted the Github repo yesterday before time but I would like to add more features that are not part of the competition. Can I do a different branch and add those feartures so only the main branch would be used for judging or can I clone the project to CodeBerg, where I was already planning to move, and continue there, or do I need to stop work until the judging is complete?

[coral granite]

I included the last commit hash before the deadline in my writeup. Still, I feel uneasy about continuing work in the same public repo. Perhaps a private fork is the safest option - unless you’re just fixing something embarrassing in the public repo?

[buoyant anchor]

I don't want to change anything in the public repo. I wanted to add some more features. I think I private fork on a different platform might be a good idea. Basically it is using one data source (IMU Data) and I wanted to add another source which would be an RGB-Depth camera data. I already mentioned that this was a planned future addition to the project.

[pliant rampart]

Thank you so much for the kind words 😭🙏

I definitely had an oversight with the rules, but yea, now is the time for aura farming in our socials 😈 😈

Anyway guys, I started a Kaggle discussion thread regarding the winner announcements to see if we can get an ETA from the judges. Feel free to check it out and follow the thread if y'all are interested in the updates
https://www.kaggle.com/competitions/gemma-4-good-hackathon/discussion/701910

[bleak matrix]

new to this server,can someone tell me when the winners will be annnounced?

[elder apex]

it would take longer than 3n i think, since 3n has 600+ subs and 4 month to be announced, 4 for the goods has 1613 its x2,7 more than 3n

[elder apex]

anyway imma ask something so if we publish latest release without commits new code and this latest release has no relevant to competition just for Public Beta safe no? or we just created new repo?

[bleak matrix]

so almost 11 months
better hope they hired some more evaluators this time

[pliant rampart]

there's 7 judges tho this time, last time is 4, and the max word counts that are not penalized is 1500 words, new rules

i didnt see the rules and make 6k words for the writeups haha, im so dead

but being optimistic, they said that it "may" be penalized so... im not dead yet ig

hopefully there's no ai involvement in the judging proses

[elder apex]

🤣 if theyre using fully ai involvement in judging were so dead ai could missed a lots technical dept it turns false positive but thats 70% ai 30% human scenario, if 50/50 like for getting new insight, make a comparison breakdown, etc. for judges not problem bc the hackathon itself has 1600+ submission

[native sail]

So, anyone please take a look on our WriteUp: https://kaggle.com/competitions/gemma-4-good-hackathon/writeups/farmwise-ai-gemma-4-good-hackathon. Please give your feedbacks on future vision of FarmWise AI. Thanks

[fickle charm]

hey everyone! saw a lot of really interesting submissions throughout the hackathon, so many amazing angles and approaches to the challenge honestly 💯

we also published our writeup here if anyone wants to check it out! would really appreciate any feedback or support from the community as well :)

https://www.kaggle.com/competitions/gemma-4-good-hackathon/writeups/wicara-adaptive-ai-tutoring-gemma

[bleak matrix]

https://www.kaggle.com/competitions/gemma-4-good-hackathon/writeups/new-writeup-1779107245532
would appreciate feedback and upvotes

[elder apex]

looks like team pcp manage to clone my private repo bc in 18 and 19 my repo still private until an hour to submission deadline it has 67 clone i wonder is really team pcp or something?

[copper spire]

probably code scrapping bot most of these AI use it for training

[elder apex]

ic ic the ai manage to scrapt in 18 has 20 clones repo still private is mindblowing, yeah prob for bitnet /ai training

[pliant rampart]

my youtube channel that specially made for this hackathon, that hosted my pitch and demo video got taken down, wth is wrong with youtube man, this is so unfair

[coral granite]

I have created a field communications kit for use under disaster conditions. I have modelled it around wildlife rescue work after cyclone Narelle in Western Australia. For on-device Gemma 4 inference, there's a dedicated Swift App (including custom LiteRT-LM bridge). And for the human-in-the-loop part, there's a Gradio 6 app. Source code, videos and test data linked in the Writeup.

Kaggle Writeup here: https://www.kaggle.com/competitions/gemma-4-good-hackathon/writeups/new-writeup-1778759643211, LinkedIn post with additional background there: https://www.linkedin.com/posts/nilsdurner_ai-4-good-from-field-evidence-to-public-activity-7464572903641137152-fKdw?utm_source=share&utm_medium=member_desktop&rcm=ACoAAAGX2jIBd6RDsNRYv13Bvu3x4nnCNu96SEw