I am interested in the layer between software and infrastructure: the part that makes systems useful, controllable and resilient. I would rather build the machinery behind the product than another thin wrapper around an API.

OpenAI-compatible gateway for teams that run several model providers. It brings intelligent routing, provider failover, policies, AI FinOps, security and observability into a single open source layer. Authentication, tenant isolation, quotas, audit and events.

Go PostgreSQL Redis Docker Kubernetes Prometheus

_{Go 87% · HTML 9% · Shell 2% · others 2%}

• Status: V5.0.1 stable · V6 in development on the v6 branch
• License: MIT
• Last activity: Oct 16, 2026

Rust post-quantum cryptographic security engine for data, keys and identities, with key lifecycle management, authenticated encryption, signatures, credentials, policy enforcement, audit logging and hardware-provider abstractions. The project includes security CI, fuzz targets, documented threat models, benchmarks and a PQC Benchmark Arena. v4.1.2 is the current release. It is an audit-ready release, not a production-readiness claim; no independent external audit or FIPS validation is claimed. The native ML-KEM/ML-DSA backend remains experimental and opt-in; RustCrypto is the default path.

Rust ML-KEM-768 ML-DSA-65 SLH-DSA RISC-V Android SDKs

_{Rust 94% · C 3% · C++ 1% · Go, Shell, PHP, Ruby 2%}

• V1: frozen baseline v1.0.0
• V2: frozen on nexusqv2 (v2.0.0)
• V3: merged into main (PR #45)
• V4: v4.1.2 current release · Audit-Ready Release · no production-readiness claim, no FIPS validation, external audit pending. Native backend remains experimental and opt-in; RustCrypto is the default.
• Companion tools: 25 independent Rust CLI tools built on the public nexusq-core API; all 25 crates are published on crates.io at 0.1.0.
• Last activity: Oct 28, 2026

A general-purpose language with a small syntax and a deep semantic model. The long-term goal is its own compiler, runtime, ABI and self-hosted toolchain, so that AXIOM compiles AXIOM. Python and Rust are the transition infrastructure.

Python Rust Compiler Runtime Systems

_{Python 51% · Rust 49%}

• Status: architecture closed, now being implemented
• License: MIT
• Last activity: Oct 3, 2026

A Rust-first foundation for runtime execution, memory, compute, ECS, deterministic simulation and early networking. Core/runtime/memory/compute/ECS foundations, five bounded simulation models, initial blocking TCP wrappers and prediction/reconciliation/replay primitives are implemented. Networking wire encoding and full protocol integration remain unfinished; this is not a production-ready runtime or game engine.

Rust Runtime Compute ECS Simulation Networking

_{Rust 100%}

• Status: pre-alpha · active development
• License: MIT OR Apache-2.0
• Last activity: Oct 19, 2026

A SKILL.md file with instructions that let an AI guess what you are thinking of, with no app and nothing to install. The questions adapt to your answers, a "not sure" counts as neither yes nor no, and it only takes a guess when it has enough evidence. It also includes a minimal remote MCP server for compatible AI clients.

Markdown TypeScript MCP

• License: MIT
• Last activity: Sep 29, 2026

An experimental project with a Python backend and a JavaScript interface, deployed on Vercel. I keep it lightweight while the implementation evolves.

Python JavaScript Vercel

_{Python 74% · JavaScript 26%}

• Last activity: Sep 28, 2026

I contribute to Arc, by Basekick Labs: an open, SQL-native time-series database written in Go on top of DuckDB, built for high-throughput telemetry, Parquet storage and analytical SQL. I have 8 PRs merged into main, covering the WAL, file replication, ingestion, retention policies, the tiering API and configuration.

Go DuckDB WAL Parquet Ingestion

• PRs merged into main: 8
• Last merge: Oct 29, 2026

Arc (Basekick Labs) is an AGPL-3.0 licensed SQL time-series database written in Go, with 686 stars. These are my changes that the maintainer has merged into its main branch, with the context of each review.

Merged into main · Oct 29, 2026 · +319 −16 · 7 files

Adds database- and measurement-filtered manual tier migrations, validates the source and destination tiers while keeping the supported direction explicit as hot → cold, and makes dry_run report candidate count and bytes without moving files. The scheduled migration path remains unchanged. The PR adds focused coverage for filtered candidates and preview non-mutation; it was merged by the Arc maintainer after review.

Merged into main · Oct 13, 2026 · +12 −6 · 2 files · Closes #1195

Removed license.enabled from the config struct, loader and defaults because it was read but never used. Enterprise licensing continues to be driven by license.file_path or license.key; any leftover license.enabled setting or ARC_LICENSE_ENABLED environment variable remains ignored, so this is a cleanup with no behavior change.

Merged into main · Retention · Cold-tier storage

Retention policies now apply to data in the cold tier as well, bringing cold-tier lifecycle handling in line with Arc's retention rules instead of limiting policy enforcement to the hot tier.

Merged into main · Oct 8, 2026 · +623 −101 · 16 files · After 26.09.3 · Closes #975 · unblocks #1092

Partial DELETE rewrites no longer mutate the existing manifest entry: they publish a fresh Parquet file, register the new path and delete the old one in a single Raft batch, with the rewritten checksum and size and the rewriting node as the new origin. Before, a same-size rewrite could leave replicas trusting a stale file, because presence was checked by size only. It covers the live, replay and snapshot catch-up paths, and serializes concurrent partial DELETEs so they cannot publish several survivors from the same source. The maintainer noted that the approach was right from the first push. Benchmark on existing data: +0.26%, within the noise margin.

Merged into main · Oct 7, 2026 · +129 −21 · 2 files · After 26.09.3 · Fixes #1135

GET /api/v1/tiering/files rejects non-integer or negative limit values, which used to cause a panic when slicing with a negative index. I added a structural guard with a positive limit at the point of use, kept limit=0 as an explicit empty result, and dropped the proposed cap of 1000 so the API stays backward compatible.

Merged into main · Oct 4, 2026 (merged Oct 5) · +137 −4 · 3 files · Release 26.09.3 · Fixes #889

Line-protocol writes reached the buffer as columns with no raw client bytes, so their WAL entry went out without a database and a receiver always filed it under "default". The row-format path now writes the same envelope that msgpack uses. It is safe for replay, receiver, checkpoints and large sizes, and readable by any binary since 26.05.1. During review the maintainer narrowed the scope: the change in cmd/arc/main.go (#886), which depended on #888, was dropped and only the #889 envelope was kept.

Merged into main · Oct 4, 2026 · +884 −156 · 8 files · Release 26.09.3 · Resolves #1009 / #966

Closes the remaining gap in WAL purging. It tracks pending identities through durable flush checkpoints, exposes a real minimum-unflushed sequence and uses PurgeFlushed instead of wall-clock age for periodic and recovery cleanup, so a file is deleted only once its data has been flushed. After review, a maintainer commit bounds the purge floor and reclaims files it cannot describe.

Merged into main · Sep 13, 2026 · +58 −2 · 3 files · Release 26.09.2 · Fixes #768

The fallback that coerces numeric values into host_<num> strings is intentional and stays, but it hid a sign of misconfigured clients. I added a debug-level log when it fires, so the hot ingest path is not flooded, and two test cases (float and boolean host) that fall through to the default branch. It went through three review rounds and the merge added me to the project contributors list.

_{Source: the public PR and release pages of Arc on GitHub and its API, as of October 29, 2026. All merges were done by the project maintainer. #769 shipped in 26.09.2, #1055 and #1056 in 26.09.3, and #1051, #1073, #1241 and #1052 were merged after 26.09.3, the latest published release.}