The Bug That Only Showed Up After a Week: Why We Still Needed to Add a Daily VACUUM
Part 3 of 3 — A Quieter Story About Ongoing Ownership
Kompress Insights - The KompressAI Tech Blog
This blog is a behind-the-scenes look at building and scaling industrial intelligence systems at Kompress.ai.
We write about what happens when software meets real machines in the field where latency, reliability, and scale are not theoretical problems, but daily constraints.
Latest articles
Part 3: Building a Cluster That Manages Itself
The partition index table, the monitoring system, and reading the sawtooth
Part 2: When 0% Tombstones Is Not the Finish Line"
Four days after hitting 0.00000% droppable tombstones on all three nodes simultaneously — the cleanest state the cluster had ever been in — the Slack report showed 30% again
Part 1: When Your Time-Series Data Turns Against You...
Migrating 3.9 Billion Rows to TWCS Without Downtime on a Constrained IoT Cluster The number that started everything was 7.99 ms. That was the average read latency on our Cassandra keyspace — a cluster
When Your Fix Creates a New Problem: Debugging the Production Partition Migration
Part 2 of 2 — Two Swap Cycles, One Rollback, and Why the Safety Net Made All the Difference
When Your Database Becomes the Bottleneck: How We Rebuilt Our IoT Telemetry Pipeline Without Taking It Down
Part 1 of 2 — The Problem, The Design, The Migration
Beyond the Model: How Interaction Design Makes AI Useful
Which LLM should you use? That’s usually the first question. But after building real-world AI systems, we realized something: 👉 The model matters. 👉 The design of interaction matters more. Most
What 365 days of telemetry data taught us about building in the real world
Most people celebrate growth. We at kompress.ai learnt to read it carefully. Over the last 12 months, we tracked a simple metric: daily telemetry requests from devices deployed in the field. This is r
