Jovan

Just another data scientist
During the day @Tiqets
At night building

[Jovan @JovanVaex]

@buildsbyaniket @emilson_tery @googlegemma This is the way! @DSPyOSS all the way. Especially with gepa, if you have a good feedback loop, it is a game changer

[Jovan @JovanVaex]

@dbreunig But maybe their spirit lives in the gemma models.. i should check that out!

[Jovan @JovanVaex]

@dbreunig Absolutely this! I share the same sentiment. I use LLMs a lot in other non coding cases where logic is needed coupled with some low/medium level of creativity. Frontier models perform worse.. which is why i am sadden for the retirement of gemini 2 and 2.5 series of models

[Jovan @JovanVaex]

@HamelHusain The 2.5 series is already scheduled to be phased out. I was super happy with the 2.0 series, and happy enough with the 2.5, but.. 3.x.. something is very off. Not talking about coding btw, other industrial usecases.. Shame really.

[Jovan @JovanVaex]

@VictorTaelin @leerob I am more and more convinced it is a pretty good model.

Would be great to have it via an api (id like to try it out on non coding tasks)

[Jovan @JovanVaex]

@theo I really miss gemini-2.0-flash-lite. i wish they would open source it. I dont what the secret sauce is, or luck, but they found it and never propagated it forward. A shame. I still use it via vertex at time, but shame you can't build on it

[Emilia @me__emi__]

I have spent a decade watching strong people feel small in front of their own careers.

People with experience, talent, resilience, and real capability. And yet, when they face a career transition, they are often reduced to a CV, a job title, or a set of keywords.

That NEVER felt right to me.

In 2016, I co-authored an IEEE published paper on technology in career services. Since then, my work has moved across Generative AI, search, knowledge graphs, product strategy, and digital transformation, now as a Doctoral Researcher at the Technical University of Munich.

Corren brings those worlds together.

Corren is a data-grounded career transition intelligence platform for people trying to understand what they can realistically become next.

The map of work is being redrawn.

AI is changing which skills are automated, amplified, combined, or made more valuable, while most career tools still treat people like static CVs.

They match keywords.

They suggest jobs.

They produce generic advice.

But they rarely explain how someone’s real experience can transfer into future roles.

Corren was built to change that.

It turns a person’s background into an explainable transition report: transferable capabilities, relevant gaps, realistic career paths, and AI-related exposure.

It is NOT an AI wrapper.

Corren is grounded in real structured career data, enriched role models, capability relationships, transition logic, and reviewable reasoning.

The goal is not to pretend that career prediction can be perfect. Career mobility depends on industry, geography, seniority, market context, regulation, personal constraints, and edge cases.

Corren is designed to surface those limitations honestly, while still giving people something many career systems fail to provide: a clearer map.

For me, Corren is a ❤️ LIFE PROJECT❤️!

It is my answer to one of the most urgent questions of the AI era: what can I become next, and what would it realistically actually take to get there?

@Replit #ReplitBuildathon #10yrbuildathon

Let's goooooooooooooooo

[Jovan @JovanVaex]

@OfficialLoganK @GoogleAIStudio

[Jovan @JovanVaex]

@OfficialLoganK Can ai studio get a little love please? There has been an incident and the gemini api has been unreliable for a week now. Some updates would be helpful. Thank you!

[Hamel Husain @HamelHusain]

I hand wrote this the slow way. Was a good feeling

[NZ ☄️ @CodeByNZ]

OpenAI's latest repo has Claude as the third top contributor 😭😂

[Yugoslavian Visuals @YugoslavVisuals]

Did you know Macedonia was this beautiful?

[Jovan @JovanVaex]

The gemini-3 preview series of models is outstanding! However, it is worrisome how often i hit the content filter for simple classification or q/a with context tasks with nothing even remotely fishy.

Hope the GA will clean this up, never happened with the 2.0 and 2.5 series.

[Jovan @JovanVaex]

cc @_philschmid

[Jovan @JovanVaex]

Just got notified that gemini-2.0-flash-lite is being sunset. It has been my go to model for so many tasks, and I've learned how to prompt it and get better performance than the 2.5 series for many cases.

@OfficialLoganK is there a hope of getting a new (series 3) lite model?

[Jovan @JovanVaex]

Given the performance of the gemini 2/2.5 series.. i was hoping that at least some would get long term support (or at least be available).

I will surely miss the 2.0 flash lite model. Unsung hero amongst models.

[Daniel Lockyer @DanielLockyer]

I love it when people work on something they're extremely passionate about

[Jovan @JovanVaex]

@aye_aye_kaplan Awesome, thanks!