TinyCubeWayEngineering notes and research
How we approach the problems we are asked about most: running streaming platforms, keeping agent context fresh, semantic layers that give the same answer every time, and connected-vehicle data. Plus a quarterly look at where Indian startup money is going.
-
Context layer
Populating an agent's context layer from Kafka, end to end
From a database commit to an agent's answer: change capture, Kafka, stream processing, a serving store and an MCP tool, with data age measured at every hop.
8 min read -
Streaming platform
Moving Kafka off ZooKeeper to KRaft without downtime
Kafka 4.0 no longer runs on ZooKeeper. A step-by-step runbook for the migration, including the parts that cause outages.
6 min read -
Semantic layer
A semantic layer on Apache Iceberg that an agent can query correctly
Why raw table access gives confident wrong answers, and what to define so that it doesn't.
5 min read -
Semantic layer
One semantic layer over scattered Postgres, MySQL and app databases
Query them where they live, or copy them into one place? A decision framework, table by table.
5 min read -
Automotive
Connected-vehicle telemetry: seven data problems that break pipelines first
Late signals, drifting clocks, duplicate uploads and firmware schema changes, and the patterns that handle them.
6 min read -
Research
India Startup Pulse, Q3 2026: where the money went, and where the moat sits
$2.9B raised and AI led. In five of the six AI-core companies we studied, the moat is not the model.
8 min read