I fix real bugs in codebases bigger than mine. Local-first tools by day, upstream bug hunting by night — every change backed by a reproduction, a regression test and a benchmark or an honest "no performance claim".

• 🧭 How I contribute: repro first → smallest correct patch → adversarial self-review → maintainer-friendly PR
• 🔨 Building: mmap-chunker-core (Rust · zero-dep · stable C ABI · on PyPI & crates.io), coreguard, aethelgard
• 🛠️ Stack: Rust · Python · C ABI · TypeScript · FastAPI · Playwright
• 📍 Berlin — building for the German market with global engineering standards
• 🤖 AI disclosure: I use AI to explore, scaffold and review — every diff is read, understood, tested and explainable by me before it ships

Merged upstream 🎉

Repo	PR	Fix
scrapy/w3lib	#291	Support quoted charset values in Content-Type headers

Open & under review (badges update themselves)

Repo	PR	What it fixes
odoo/odoo	#283999	isolate defaults when creating relational values via import
odoo/odoo	#283316	preserve error cause + transaction semantics in QWeb t-nocache
odoo/odoo	#281986	normalize overtime day start in hr_attendance
pydantic/monty	#768	clean up object keys when value parsing fails
scrapy/w3lib	#306	keep percent-encoded % in canonicalize_url
langfuse/langfuse-python	#1826	prevent stale clients queueing after shutdown
pretzelai/pretzelai	#171	keep allow_hidden when extensions wrap the contents manager
OpenHands/automation	#283	skip cron runs while an automation is active
OpenHands/automation	#277	purge expired local-mode run workspaces
opensanctions/rigour	#261	LangStr hash/eq contract violation (closes #255)
alunduil/zfs-replicate	#512	Vale style rule for agent instruction files
kubernetes/kubernetes	#141338	DRA: malformed per-device node selectors as node failures (draft)

Also: signed Odoo Individual CLA · reported OpenHands/automation#278 (dispatcher-owned tasks not drained at shutdown)

Every PR above has a reproduction and a regression test. No drive-by cleanup, no AI slop, no contribution farming — one solid fix beats ten speculative PRs.

Project	What it does	Badges
mmap-chunker-core	Record-aligned byte-range planning, zero-copy mmap, stable C ABI — CLI + Python wheel + Rust crate
coreguard	Windows-native C execution layer — Job Object containment, resource limits, metrics
aethelgard	Local-first NIS-2 evidence triage — Article 21(2) matrix, SBOM, audit ledger
algorithm-evolver	Self-evolving algorithm experiments
atlas-mvp	Early-stage mapping / MVP experiments
mev-sentinel	MEV / market microstructure sentinel

def ship(change) -> str:
    assert reproduces_real_bug(change)      # repro before fix
    assert minimal_diff(change)             # understand broadly, change narrowly
    assert tests_prove_it(change)           # failing test first, green after
    assert i_can_explain_every_line(change) # AI-assisted ≠ AI-authored
    return "merged or honestly closed — never farmed"

• Proof before claim: pytest -q · cargo check · benchmarks — no perf claim without a measurement
• One issue = one worktree: focus over context hopping
• Maintainer time is expensive: small diffs, clear repros, zero hand-holding required
• AI is a tool, not an author: I prompt, read, verify and take responsibility

“Repro first. Smallest correct patch. Proof over promises.” — rufus
_{Animated with capsule-render • typing-svg • skillicons • stats cards • activity-graph • snake • live shields.io badges}