Projects

Two of these are in active development and I write about them as they change. The rest shipped and run in production, or ran long enough to matter.

Active

ursus

v0.2 · active development

A dataframe library for Go, modelled on Polars

Lazy execution with a query optimizer, Arrow memory layout, SIMD kernels, and streaming execution that spills to disk rather than falling over. Pure Go — no cgo, no C++ toolchain, no Python runtime.

22/22 TPC-H · 15/15 h2o.ai, validated against DuckDB
1,610 tests across four SIMD widths
Pure Go — no cgo, static binary

logd

Phases 1–9 complete · active development

A log storage engine with typed-range indexing

A single-node log daemon with a Loki-compatible API. Its differentiator is range queries on typed values pulled out of unstructured log text — answer duration > 200 in O(log n) without pre-structuring the data.

Loki-compatible: Grafana points straight at it
O(log n) typed range queries over raw log text
Also my MSc thesis, under prof. Krzysztof Stencel

Shipped

Real-time aggregation pipeline

OOM restarts → zero

Go orchestrator, short-lived Python workers

A 24/7 aggregation service on 2 CPUs and 3 GB of RAM. Migrated from Pandas to Polars for a ~50% speedup, then re-architected into a Go orchestrator with short-lived Python workers to kill a memory leak I could not find.

Telegram bot

100K+ users

Webhook handling to data storage, end to end

Built and operated a high-traffic Telegram bot end to end — webhook handling, rate limits, state, and data storage — serving over 100,000 users.

Logistics CRM

In production use

Full digitalization for a logistics company

A Django CRM that digitized a logistics company’s whole back office — accounting, income, invoices, reports, employees, salaries and KPIs. Still running the business day to day, and still mine to maintain.