Skip to content
View izman48's full-sized avatar

Block or report izman48

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
izman48/README.md

Hi, I'm Iesa 👋

How to work with me

Plenty of people keep a short "how to work with me" manual — a user guide to themselves. This is mine. It's a shortcut for teammates, mentees, candidates, and anyone collaborating with me on a project, written in my personal capacity and reflecting how I like to work rather than the position of any organisation I'm associated with.

Don't treat this as a rulebook. It's a shortcut to getting on the same page faster. Ignore anything that doesn't fit the situation.

From you to me

  • Lead with the ask. Tell me what you need and by when, up front. If there's an action for me, make it unmissable — put it in the subject line or the first sentence, not buried three paragraphs down.
  • State your uncertainties and assumptions. I'd much rather you flag "I think X but I'm not sure" than present a guess as a fact. It tells me where to look.
  • Bring a goal, not just a problem. "I want to achieve X, I've tried Y, I'm stuck on Z" gets a far better response from me than "this is broken."
  • Unpolished and timely beats polished and late. Send me the rough draft, the half-formed idea, the early warning. I'd rather course-correct on day one than admire a finished thing going the wrong way.
  • Prefer async by default. A clear message (Slack, email, a doc, a PR comment) is usually better than an unplanned call — it gives me time to think and gives us both a written record. Exception: if you're blocked and I'm the blocker, interrupt me.
  • For anything substantial, write it down. A short doc or ticket beats a long thread. It's easier to review, comment on, and find again later.
  • Default to communicating in the open (channels, not DMs) so others can learn from it later — unless it's personal or sensitive.

From me to you

  • I'll ask a few clarifying questions, then get moving. I won't interrogate you, but I'll usually check the goal and the constraints before diving in. Vague requirements are where projects quietly go wrong.
  • I work in small iterations. Expect me to ship something rough early and refine it with you, rather than disappear and return with a finished artifact. If I'm heading the wrong way, the sooner you tell me the cheaper it is to fix.
  • I optimise for the code that's still readable in two years. I'll favour maintainability and clarity over cleverness, and I'll push back (kindly) on shortcuts that we'll pay for later.
  • I write tests first where it makes sense. If you see me building the test before the feature, that's deliberate — it keeps the design honest and the scope tight.
  • Assume misunderstanding, not malice. If something I said landed badly, it's almost certainly clumsy wording, not intent. Tell me and I'll fix it.
  • I don't always send acknowledgement messages. No reply for a bit usually means I'm heads-down or overloaded, not that I'm ignoring you. Nudge me if it's time-sensitive.
  • I'll give you honest, direct feedback — and I want the same back. Feedback is a gift, even the uncomfortable kind.

How I think about the work

  • Goals first, then sprints. I like a clear objective and short, visible cycles of progress against it. Momentum compounds.
  • Maintainable above all. Readable, well-tested, boring-in-a-good-way code beats a clever one-liner nobody can touch.
  • Tight feedback loops. The faster we learn we're wrong, the better everything gets.

How to reach me

  • Discord: izman48 — best for a quick message or to start a conversation.

Pinned Loading

  1. llm-control-plane llm-control-plane Public

    Toy-scale but honest LLM inference control plane: router + autoscaler + live observability over continuous-batching workers.

    Python

  2. llm-rl-playground llm-rl-playground Public

    Tiny RL environments + graders + reward-hacking QA for LLMs: a verifiable code track (RLVR) and an open-ended truthfulness track, plus an MCP server. Pure-stdlib core.

    Python

  3. 2d-fighting-game-rl-ai-training-tool-dissertation 2d-fighting-game-rl-ai-training-tool-dissertation Public

    My university dissertation where I trained an AI using RL to become better at a custom game i built

  4. finance-tracker finance-tracker Public

    A dashboard i uses to track my finances. This uses truelayer for the open banking api to automatically pull bank data. Trying to answer the questions "Where did my money go?", "How much money do I …

    Python

  5. jnw jnw Public

    A website to advertise vehicle fleets and tourism services across sri lanka

    TypeScript

  6. iesawazeer.com iesawazeer.com Public

    Personal Portfolio

    Vue 1