Ali Akhtari

Ali Akhtari

I build a lot of things. I ship the ones that earn it.

Lately that's mostly AI — small language models, agents, and whether you can actually trust what they produce — with a long-running streak of security and reliability work. SpotifyScraper is the one that found real users; the rest is an honest notebook of experiments, half of them unfinished, all of them teaching me something. Wander in.

See the work Say hello
A small map of this site An interactive map linking to the work, about, now, notes, and contact pages. Work About Now Notes Contact

Each point is a page. Hover to look closer.

The map has another side.

The other side · AI-assisted

The map, speaking.

Ask about the work, the approach, or what has his attention now. The voice is narrow by design: it answers from approved public material and points back to real pages.

  1. Page

    The public map is awake. Ask a useful question and I will point to what I know.

Complete the quiet verification before asking. Each response is generated when you submit.

This is a scoped AI interaction. Submitted questions are sent to OpenAI for a response. It is not Ali, and it does not have access to private information.

01 The one that shipped

SpotifyScraper — the one people actually use.

A Python library that pulls clean, typed data out of Spotify's public web player — tracks, albums, artists, playlists, and podcasts — without official API keys. It isn't a demo or a placeholder: it's a real package on PyPI with real users, a documentation site, and the unglamorous maintenance that keeps a library alive across breaking changes. If you want to see how I build and support something other developers depend on, this is the honest example.

2026 · active · open source

SpotifyScraper

A Python library that turns Spotify's public web player into typed data — without official API credentials.

PythonWeb scrapingOpen sourceAutomation
Read the project details
259 Stars
28 Forks
303 Commits

02 Everything else

The rest is a workshop.

Behind that one finished thing is a long bench of half-built ones — AI tools and agents, web platforms, automation scripts, the occasional native app, and a steady undercurrent of security work. I build a lot, and I ship a little: the things that earn it get finished, and the rest become lessons for the next attempt. I've made peace with that, because the unfinished bench is where most of the learning happens.

A few of the public signals are real, though, and worth being honest about. The numbers below are fetched live when the site builds — never invented, never rounded up.

259 GitHub stars
1,312 SO reputation
32 Answers shared
25 Questions asked
More about how I work

03 Now

What has my attention.

  • 01Learning small language models properly — what they can really do at sizes you can afford to run, and how to tell when one is actually any good.
  • 02Building agents with guardrails I trust: structured output, moderation, rate limits, hard cost caps. (The little 'ask' on this site is one of them.)
  • 03Treating verification as a feature, not an afterthought — proving that model output is faithful to its source before anything depends on it.
The current page

A good place to begin

Working on something interesting? Tell me about it.

Say hello
boiler room — ali@aliakhtari.com