A Quick rundown on concurrency and gc
How Go, Kotlin coroutines, and BEAM handle preemption safely, why GC safe-points exist, and a source-grounded tour of tricolor GC, G1, ZGC, Shenandoah, C4, and more.
iko’s logs
Yet another platform engineer who'd rather design systems, single malts, and write tools than glue together someone else's API.
Latest posts
Optimux: from a worker-pool tutorial to a DynamicScaler
The Job/Dispatcher worker pool from the famous '1 Million Requests' article is built for I/O-bound fan-out. Image processing is a mixed I/O+CPU pipeline, and pprof proved the difference. Here's the road from that pattern to a self-scaling worker pool.
CodeKeyboard Rebuild Notes
Written so a human or an LLM can rebuild the keyboard without inventing APIs. CodeKeyboard is a system Android keyboard (IME), not a text box demo — split Sofle-inspired layout, multiple layers, home-row modifiers, and on-device prediction that learns personal and mixed-script typing (Banglish included) without a server. These notes walk through the canvas rendering, composing/selection handling,…
Shrinking a Materials Database from 11GB to 266MB Without Losing the Answer
materialsDB is a small Go API serving 145,923 materials from the Materials Project, backed by a single SQLite file. Live at materials-db.fly.dev.
Chasing 1 Million RPS
yes, this is clickbait. kind of.
Materials Science Engineering Notes
My brief stint at a materials science company. These are technical notes on the engineering behind a research platform — agent architectures, hypothesis-driven workflows, and document processing pipelines for scientific papers.
Learnings from Building LLM Systems
My previous employment was at a company in the Digital asset management space, working on making short form commercial videos, towards the end.
Domain is Moat
Most teams treat prompts and models as the product. The real moat is understanding the domain deeply enough to build the right decision systems and feedback loops.
RUNE
Building a wrist gesture controller for smart devices — TV, bulbs, plugs, desktop. Not a fitness tracker.
PostgreSQL Internals Deep Dive
A wizard-level exploration of the PostgreSQL codebase, from SQL string to disk blocks and back.