Skip to content
View arbazsiddiqui's full-sized avatar
🏡
Working from Home
🏡
Working from Home

Block or report arbazsiddiqui

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.

Content in all repositories owned by your account will be closed.
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
arbazsiddiqui/README.md

Backend engineer, ten years, mostly consumer products at scale. Principal Software Engineer at Turnip, looking after the platform behind 25M users and the AI infrastructure sitting on top of it.

The rest of the time I build my own things, from training the model to shipping the app.

Models

  • VoxLock · Keeps the caller's voice and removes the other voices in the room before speech-to-text hears them. The lowest word error rate on the Krisp Voice Isolation Benchmark of every model tested, paid or open. Streams on CPU, works as a Pipecat audio filter, and runs in your browser: try it on your own voice. Code · pip install voxlock
  • kev-0.6b-browser-use · A 0.6B decision model for browser agents that picks the element and the action in one forward pass. 32.2% step success on Mind2Web, level with Kev-9B at a fifteenth of the size, and it runs in the browser on WebGPU. Code · Write-up
  • Ozan-v1-12B · An open creative-writing model fine-tuned from Mistral-Nemo on $40 of GPU. Lowest slop score of any 12B measured on EQ-Bench. #1 trending on Hugging Face, 3K+ downloads. Hosted on Featherless, try it with no download. The training recipe is public.

Apps

  • Iris · A cycle tracker for iOS. Every feature free, twelve languages, health data never leaves the phone. 4.9 stars, 1,000+ downloads.
  • Mercuro · Thermometer logic puzzles running on their own constraint solver. Native on iOS and Android. Featured in gaming newsletters.
  • Glass Jar Garden · A terrarium simulator in the browser. Build a jar layer by layer, plant from 90 real species, seal it and watch it live day by day.

Pipelines that run themselves

  • The Footnote · A history channel that researches, scripts, voices and publishes itself. 500K+ monthly views.
  • Iris growth pipeline · The app markets itself. Posters generated, checked against a written creative contract, published daily. 100K+ monthly views driving organic traffic.
  • Animated Hindi horror · A model writes the screenplay, image models hold the cast across 25 shots, and Wan 2.2 animates it on rented GPUs. Under a dollar a film.

Open source

  • sqlite-sparse · Semantic search inside a SQLite file. Documents are encoded once at insert; queries need no model and no server, and take 3.1 ms across a million passages. pip install sqlite-sparse
  • lru-cache-node · O(1) LRU cache for Node, with expiry and stale reads.
  • anabranch · HTTP load balancer and reverse proxy in Go.

arbazsiddiqui.me · Hugging Face · LinkedIn · Writing

Pinned Loading

  1. Ozan Ozan Public

    Ozan — an open, reproducible pipeline for a low-slop long-form creative-writing LLM (Mistral-Nemo-12B finetune). Measured before/after slop deltas.

    Python 4 1

  2. voxlock voxlock Public

    The best performing voice isolator there is. Keeps the caller's voice and removes the other voices in the room before speech-to-text hears them. Open source, streaming, runs on CPU and in the brows…

    Python 33 3

  3. sqlite-sparse sqlite-sparse Public

    Semantic search in one SQLite file. No model, no server at query time.

    C 13

  4. kev-browser-use kev-browser-use Public

    A tiny 0.6B Jev-like model fine-tuned from Kev for browser use: 32.2% step success on Mind2Web, runs in your browser on WebGPU.

    Python 3

  5. lru-cache-node lru-cache-node Public

    A lighting fast cache manager for node with least-recently-used policy.

    JavaScript 119 8

  6. anabranch anabranch Public

    A simple HTTP load balancer and reverse proxy written in Go.

    Go 28