Computer science student. I prefer to state what a problem is and where its constraints lie before talking about tools — the conclusion often differs from the conventional answer. What I have built is modest, but each piece is meant to be reproducible, verifiable, and open for someone else to continue.
The answer depends on real limits, not on industry habit. Setting labels aside and reasoning from the constraints often leads somewhere else.
From interface and data to inference and packaging, I go through each layer myself — not to prove breadth, but because crossing every layer is how you find where a problem actually sits.
Tests, baselines, mutation smoke checks: confirm the gate turns red before trusting its green. Stability comes from repeated triggering, not from a few observations.
When a framework misbehaves, I first separate my misuse from a real defect. If it is the latter, I reduce it to a minimal reproduction and report it upstream.
Two gaps in the official async docs; also led to a framework fix in #7106.
run_task() exceptions were logged under
concurrent.futures rather than flet, so a
flet-only filter missed them entirely.
Page.update() thread-safety was undocumented;
run_thread() does not return to the UI thread.
The 1.x scheduler calls asyncio.Event.set() across threads
unsafely. Independently reproduced and cross-checked against the fix.
run_thread() workers could not be cancelled; the process
lingered after the window closed. Minimal reproduction with a clean venv
and a no-worker control group.
A semester timetable tool for HarmonyOS PC. Detects four classes of scheduling conflict and proposes actionable fixes.
Scanned 141 controlled files with ScanCode Toolkit 32.5.0, checking Apache-2.0 against LICENSE / NOTICE / file headers, including false positives and blind spots.
FDE coursework covering business discovery, solution design and a runnable prototype for handover defect tracking and rectification.