Allegro Tech

20/08/2026

How do you evaluate search relevance at scale? 🧐

We built an LLM-as-a-judge framework grounded in human annotations. Our best local model matched cloud quality while cutting costs by 60%.

Read more on : https://blog.allegro.tech/2026/08/automating-search-relevance-llm-as-a-judge.html

11/08/2026

Back to the data center in 2026 (just a little bit) - read our story of why cloud isn't always the answer.

We learned this the hard way when a pipeline bug caused our service to repeatedly download 10GB of files from Google Cloud Storage, burning $1,000 in egress costs in just two hours. 💸

Discover how we identified the problem, considered multiple solutions, and ultimately moved back to our data center using Ceph S3-compatible storage to avoid similar incidents in the future.

A practical look at making the right architectural choices for your infrastructure on ➡️ https://blog.allegro.tech/2026/08/back-to-dc-in-2026.html

10/08/2026

The dust may have settled and the music stopped, but the Pol’and’Rock energy will stay with us for a long time. 🎶🎶🎶

As is tradition, Allegro created a dedicated backstage space for our team members who dropped by in their free time to recharge, catch up, and just enjoy the vibe together. 🎡 Many of them were also behind the scenes, organizing our festival presence and making sure you had an unforgettable experience! 🥰

Take a look at how it went! 🧡

04/08/2026

experts on stage? Of course we like it! 🤩

Michał Junczyk - Senior Manager, Data Annotation Services - was a speaker during the GenAI in Localization Conference 2026 and we would like to invite you to watch the recording of his speech.

The recording of the talk, "Evaluating Agentic AI: Why It's Hard, and Why Language Teams Already Have the Answer," is waiting for your views! 👀

➡️

We are going LIVE with the biggest AI + Localization event of 2026:...

03/08/2026

How we achieved independent deployments in a micro-frontend architecture? Read on ➡️ https://blog.allegro.tech/2026/08/how-we-achieved-independent-deployments-in-a-micro-frontend-architecture.html

At , more than 120 development teams contribute to opbox-web, the service that renders our entire frontend, currently pulling in almost 1200 npm dependencies. For a long time, shipping changes meant relying on twice-daily "release trains": gathering every author, watching for errors, and blocking everyone if a rollback was needed. 👀

In this post we walk through how we moved from that model to independent deployments: any new frontend component version can be enabled or disabled per environment at any time, without redeploying the whole service. We go into the architecture behind it and the full deployment flow, from publishing a package to rolling it out to 100% of users.

If you're dealing with a large-scale micro-frontend setup, or just curious how a monolithic rendering service can support fully independent releases across many teams, this one's for you.

29/07/2026

At , we don’t ask "who AI should replace, but what". ✨ We want our engineers to focus on what they do best, rather than getting bogged down in tasks that don't help them grow.

How do you build a mature, AI-supported engineering environment at Allegro's scale? Miłosz Kopij (Senior Manager, Product & Engineering) breaks it down in his latest interview with justjoin.it ➡️ https://justjoin.it/blog/milosz-kopij-allegro-ai-inzynierowie

27/07/2026

How AI agents halved the effort of migrating 500 Gradle modules from Koin to Hilt ➡️ https://blog.allegro.tech/2026/07/how-ai-agents-halved-effort-migrating-gradle-modules-koin-to-hilt.html

Cold start time is a retention killer, and for the Allegro Android app, Koin's runtime dependency injection was one of the biggest culprits - consuming up to a third of startup time across nearly 500 Gradle modules.

This post walks through why we moved to Hilt (compile-time DI generation instead of runtime resolution), and how we're pulling it off without freezing feature delivery: a core-first migration order, backward-compatible proxying via EntryPoint, and temporary KoinComponent glue code so migrated and non-migrated modules can coexist.

The most interesting part might be the productivity data. We ran a crossover experiment comparing fully manual module migrations against AI-agent-assisted ones, controlling for individual developer skill. Result: active developer time dropped by about 76%, and total elapsed time (including autonomous agent work) dropped by roughly 53%.

If you're dealing with large-scale DI migrations, Android startup performance, or figuring out where AI coding agents actually pull their weight versus just look good in a demo, this one's worth a read.

21/07/2026

A powerful boost of energy, practical insights, and inspiration - that’s the best way to sum up our recent internal Women in Tech meetup at . This time, we gathered to talk about topics that resonate with every one of us: the courage to be yourself in business, the power of hard data, and driving real impact within the organization.

🔸 Karolina Kozłowska (our VP, Technology) shared her career journey, reminding us that lasting confidence is built on solid expertise and staying true to your values - allowing yourself to be authentic rather than trying to fit into someone else’s script.

🔸 During her , Tina Sobocinska brought concrete data to the table, proving that diversity is also a driver of better financial performance, not just a PR add-on. This naturally led into a panel discussion moderated by Patrycja Miller (our Director, Product & Engineering).

🔸 Agnieszka Fedor, Tina Sobocinska, and Alina Magowska (our Experience Design & Insights Director) discussed how to smartly translate the EU’s "Women on Boards" directive into everyday leadership practices, ensuring that regulations are backed by a genuine culture of inclusion.

The event wrapped up with networking that carried these conversations into the halls, showing the incredible strength of our community when we connect openly, authentically, and with mutual support.

Take a look for yourselves!

20/07/2026

One speaker, two events! 🎤

You might have run into Mateusz Nalepa (our Senior Software Engineer) at DevOps Days Kraków, where he gave a Lightning Talk titled "Response times in services may not tell us the truth". 🔍 But that’s not all - he also appeared at Toruń Java User Group meetup with his presentation "Thread pool and queue wait time: A bottleneck that even AI didn’t find". 💥

As far as we know Mateusz, this was not his last word on the industry stages this year! Stay tuned for more! 🔥

15/07/2026

Is it possible to spend seven years at one tech company and never feel bored? 🔥

Wojciech Kasperski, currently a Senior Front-End Software Engineer at , is a great example of this. His Allegro journey began with a hackathon victory and an internship, as we wrote about in Just Geek IT in 2019. Today, Wojtek returns to discuss why cutting "elephant" into pieces is an important engineering skill and why, seven years later, Allegro is still a great place for him to grow.

Read the full article at: https://justjoin.it/blog/7-lat-w-allegro-wywiad-kasperski