Sovereign AI consulting for critical operations. We build and run it with Lascaris, our own platform on Kafka. IBM Partner. Malta and Miami.
Malta, EU
Joined December 2020
- Tweets501
- Following57
- Followers52
- Likes304
Sunday Kafka tip.
Before you scale out consumers on a busy topic, check its partition count. One consumer group can't have more active consumers than the topic has partitions. The extra ones just sit idle.
How do you pick the partition count for a new topic?
Weekend question for Kafka people.
When did you last test a restore? Not the backup job. The restore.
If the answer is "not sure", that is the gap auditors find first.
scalytics.io/kafshield
One fund. Two books.
Tokens on chain, the legal register off chain. Kerovan keeps them in step and proves every break.
scalytics.io/kerovan
DORA tip: Article 12 asks for backups production can't change, segregation, and tested restores.
Check one thing today: can your production credentials reach your Kafka backup? If yes, it fails segregation.
scalytics.io/lascaris
Before your first AI agent goes live, answer 4 questions:
1. Who stops a bad action?
2. Can you show why it acted?
3. Where does your data go?
4. Can you leave?
Our answers: scalytics.io/ai-agents-on-ka…
Can you trace what your agents saw when they got it wrong?
scalytics.io/check
Your Kafka is replicated. Is it backed up?
Delete a topic and your standby deletes it too. Two clusters, no data.
4 tests a real backup must pass: scalytics.io/kafka-backup-vs…
A bug ships at 09:14 and corrupts events for 47 minutes. Your standby is corrupted too.
With Lascaris you restore the topic as of 09:13 into a new topic, check it, then move consumers over.
scalytics.io/kafshield
Every AI agent action should answer 4 questions:
What did the source say?
What is true enough to decide now?
May this decision be made here?
What changed, and who allowed it?
IBM + Confluent Kafka is where those answers live: scalytics.io/blog/ibm-conflu…
Myth: our Kafka standby is our backup.
Fact: a standby mirrors your mistakes. A misfired topic delete or a leaked producer key reaches it too.
A backup needs a locked copy prod credentials can't touch:
scalytics.io/kafshield
Agents that act can also act wrong.
Lascaris runs every action through Sense → Decide → Check → Record on your Kafka. Policy first. Full audit log. Zero bytes leaving your network.
scalytics.io/ai-agents-on-ka…
Your AI agent just blocked a payment.
Can you show what it saw, what it decided and which rule allowed it?
Chat logs are not an audit trail. How we do it: scalytics.io/ai-agents-on-ka…
Giving AI agents direct database and API credentials creates unmanageable risk. Learn how to authorize capabilities safely using an event-driven Kafka pattern, Read the full post at scalytics.io/blog/agent-capa…
Traditional ROS-to-Kafka bridges lose operational context. Discover why Lascaris provides the edge-adjacent backbone needed for governed autonomous fleets. Read the full post at scalytics.io/blog/why-lascar…
Copying live data just to run AI? You're paying a massive storage tax.
We built a 45-second architecture diagnostic for Kafka and Confluent teams.
No fake ROI calculators, just a hard score on your data-movement bottlenecks.
Test your stack: ai-ready.scalytics.io/?=5
Connecting an AI agent to Kafka via MCP provides data access but lacks a governed runtime. Learn why autonomous systems need durable decision state to act safely. Read the full post at scalytics.io/blog/why-kafka-…
Giving an AI agent the right facts does not mean you control what it does next. Because prompts rely on hope rather than enforcement, production AI requires strict execution boundaries.
scalytics.io/blog/grounded-w…
Deep search AI for operations needs current evidence, active policy, and replayable decisions. This essay explains when search is enough and when a stream-edge runtime is required. Read the full post at scalytics.io/blog/ai-isnt-ha…
AI fabric is not another data catalog. It is the runtime that keeps context, policy, lineage, and replay attached to each AI decision, so actions stay explainable after the demo ends.
Read:
scalytics.io/blog/ai-fabric-…