kloia

kloia Simplifying software and infrastructure management, helping customers to focus their core-business. DevOps&Micro Services Consultancy

13/05/2026

Three days. Constant rain. Half the activities crossed off the list. Still one of the best offsites we've had.

When you work distributed, some in different cities, some abroad, all in different Slack timezones, you forget what it's like to actually be in the same room. As usual though, loud. Opinionated. Funnier than expected. Hard to replicate on a Slack call.

We talked about work, where the industry is heading and where we're probably wrong about it. Shared what we've been building, AI projects, new technologies, what worked and what didn't. We played games that got oddly competitive for a group of people who debug production systems for a living. We sat around and did nothing useful and that was somehow the most useful part.

The rain cancelled the outdoor stuff. Nobody really cared.
See you at the next one. Bring a jacket.

We call it an AI Gateway. You can call it “the thing we should have built six months ago.”Three providers, five accounts...
07/05/2026

We call it an AI Gateway. You can call it “the thing we should have built six months ago.”

Three providers, five accounts, one invoice that explains nothing.
Idle endpoints billing whether traffic flows or not.
ROI measurement? Next quarter. Still.

One gateway, every cost attributed, no invoice surprises.

We start with an AI Cost Assessment.
If the problem is smaller than you think, we’ll be the first to say so.

kloia.com/genai

Is MCP really the answer for AI agents, or did we just buy into the hype? 🤔At kloia, we don't blindly follow the hype. W...
05/05/2026

Is MCP really the answer for AI agents, or did we just buy into the hype? 🤔

At kloia, we don't blindly follow the hype. We try things, we break things, and we share what we actually learned.

That's exactly what our AI Champion Ata Ağrı will be doing at AWS Community Day Türkiye 2026 with his talk:
✨ "MCP was a mistake. Bash is better. Or is it?"

📅 9 May 2026
🕐 13:30 - 13:45 | Track 2
📍 Wyndham Grand Istanbul Levent

Come challenge the take, or get challenged. See you on Track 2! 🚀

Your RAG pipeline is hallucinating. Adding more vectors won't fix it.Vector search was built for semantic similarity. No...
29/04/2026

Your RAG pipeline is hallucinating. Adding more vectors won't fix it.

Vector search was built for semantic similarity. Not for "who reports to the person who approved invoice X." Four ways it quietly breaks in production:

Multi-hop facts get split across chunk boundaries and never come back together
Logically distinct entities cluster too close in embedding space
No awareness of when a fact went stale
Acronyms and product codes carry no semantic signal
Knowledge graphs handle relational queries natively. GraphRAG (Microsoft, 2024) was the first serious attempt to combine the two. LazyGraphRAG then cut the indexing cost by up to 90%.

We broke down the trade-offs between knowledge base and knowledge graph for teams past the demo stage. Worth a read.

Where does your RAG break first: multi-hop questions or stale data?

https://www.kloia.com/blog/knowledge-base-vs-knowledge-graph-llm

Nobody remembers the sponsor banner.They remember the conversation at the booth. The talk that reframed how they think a...
22/04/2026

Nobody remembers the sponsor banner.

They remember the conversation at the booth. The talk that reframed how they think about a problem. The person they met who’d been dealing with the exact same mess.

That’s why we’re at AWS Community Day Türkiye as Platinum Sponsor on May 9.

Not to put our logo on a wall. To be in the room where AWS engineers, cloud practitioners, and DevOps teams actually talk shop, about EKS, agentic workflows, AI-driven SRE, and the infrastructure decisions that keep people up at night.

Kloia will also be on stage. Come find us.

Most Kubernetes releases are just changelog theater.New alpha features nobody runs in prod. Deprecations that break your...
20/04/2026

Most Kubernetes releases are just changelog theater.

New alpha features nobody runs in prod. Deprecations that break your weekend. Announcements dressed up as progress.

1.36 breaks the pattern. Not with features. With honesty.

No flashy additions. No rebranding. Just a team that looked at a decade of accumulated pain and actually did something about it.

MutatingAdmissionPolicy is now GA. That's a whole class of webhook servers you can finally kill. SELinux mount relabeling now means millisecond pod startups instead of second-long hangs. Not a headline. Just a fix that was overdue by about three years. Job suspension that actually lets you change resources mid-run? ML teams have been duct-taping around this for ages.

The boring releases are the ones that respect your production environment.

But here’s what I keep wondering, how much of your current cluster complexity exists only because a previous release forced you into it?

How much of your infra is just accumulated workarounds?
👉 https://www.kloia.com/blog/kubernetes-1-36-whats-coming

"State-of-the-art" AI shouldn't result in "end-of-budget" infrastructure.To be clear, the problem usually isn’t the AI. ...
15/04/2026

"State-of-the-art" AI shouldn't result in "end-of-budget" infrastructure.

To be clear, the problem usually isn’t the AI. It’s the infrastructure debt.
Fixed capacity and manual management kill budgets way faster than bad models do.
So we documented how we built our GPU-as-a-Service architecture.
We’re showing how we solve for scaling and reliability on OpenShift and AWS.
No budget bleed.

Technical breakdown here: https://www.kloia.com/blog/gpu-as-a-service-architect

What’s your biggest infrastructure headache right now?

Making tech work is a mess, and that is exactly why we should talk about it tomorrow at the Tech Leaders Summit. 🎯As we ...
14/04/2026

Making tech work is a mess, and that is exactly why we should talk about it tomorrow at the Tech Leaders Summit. 🎯

As we head to Istanbul to join the people driving Türkiye’s digital transformation and AI, it is clear that buzzwords do not solve problems.
Our goal for tomorrow is to skip the hype entirely.

While leadership and efficiency look great on a slide deck, pulling them off in the real world is a headache.
Therefore, we are focusing on the reality of implementation rather than just talking about it.
And yes, keynotes certainly have their place, but the best insights usually happen in the hallway.
A 10-minute talk about how a problem was actually solved often stays with you longer than any formal presentation.

If you are at Radisson Blu Şişli tomorrow, let’s connect and talk shop. ☕

We just finished the APAC Team Summit. It was a lot in the best way possible. 🌏We spend most of our year collaborating a...
09/04/2026

We just finished the APAC Team Summit. It was a lot in the best way possible. 🌏

We spend most of our year collaborating across time zones, which works fine. But nothing quite beats sitting in the same room to solve a problem. It turns out that ideas move a lot faster when you do not have to wait for a calendar invite.

Here is what we actually did:
🔵 Got on the same page: We made sure every market is moving in the same direction.
🔵 Built things fast: We created new frameworks on whiteboards instead of slide decks.
🔵 Actually connected: We reminded ourselves that we are a team, not just icons on a screen.

Swipe through to see what we stayed up talking about. 📸 ↓

To everyone who made the trip: thank you for the energy and the honest conversations. More to come soon. 🚀

We're excited to be at Future of CIO Summit 2026 as a sponsor, April 9, Cıragan Palace Kempinski, Istanbul.500+ technolo...
07/04/2026

We're excited to be at Future of CIO Summit 2026 as a sponsor, April 9, Cıragan Palace Kempinski, Istanbul.

500+ technology and business leaders gathering to explore what leadership looks like in an AI-driven world.

Come find us at our booth. We'd love to connect, exchange ideas, and have a real conversation.

See you there. 👋

Control Amazon Bedrock Token Usage at Scale with Zero-Latency Guardrails 🚀Scaling Generative AI within a team usually cr...
30/03/2026

Control Amazon Bedrock Token Usage at Scale with Zero-Latency Guardrails 🚀

Scaling Generative AI within a team usually creates a direct conflict between developer speed and budget safety.

Tracking individual token consumption becomes a manual burden, and most teams try to solve it with an API Gateway proxy.

Then they hit a wall. The 29 second hard timeout of a proxy kills long running model responses and forces you to refactor your code.

We built bedrock-token-cap as a fully serverless enforcement layer ☁️

How it works (The 4-Step Process):

🔁 EventBridge triggers a Lambda every 10 minutes
📊 Lambda queries CloudWatch Logs for real-time token usage
🗄️ DynamoDB stores and updates usage data per user
⛔ IAM Deny Policies automatically block users who hit their limit

The Benefits of this Architecture:

✅ Zero Latency: No proxy slowing down your model inference
✅ Zero Refactoring: Developers keep using their existing SDKs and workflows
✅ Automated Guardrails: Token limits are enforced the moment a user hits their quota

Give your team the freedom to build and your CFO the peace of mind they need 💡

Explore the open source repo: https://github.com/kloia/bedrock-token-cap

Amazon Web Services

Address

İçerenköy Mahallesi Umut Sokak Quick Tower Plaza Ofis Sit. No: 8-10D/5 Kozyatağı/Ataşehir
Istanbul
34752

Telephone

+902162258382

Website

Alerts

Be the first to know and let us send you an email when kloia posts news and promotions. Your email address will not be used for any other purpose, and you can unsubscribe at any time.

Contact The Business

Send a message to kloia:

Shortcuts

Share

Category