Skip to content
View h3nock's full-sized avatar

Highlights

  • Pro

Block or report h3nock

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
h3nock/README.md

hi, i'm Henok👋

i like writing code and building software.

lately that has been mostly developer tools and ML systems: terminal tools, native apps, local AI/search, and training models.

recent projects

  • remux: native iOS client for remote tmux workspaces over SSH. It maps tmux sessions, windows, and panes into native mobile UX.

  • kbolt: local retrieval engine for notes, docs, and code. It indexes folders and supports keyword, semantic, reranked, and deep search, plus MCP access for agents.

  • ai-deep-dive: interactive learning platform for language models, with lessons, visualizations, coding challenges, and a backend judge for running tests.

  • codex-meter: fast tui dashboard for Codex usage, quota, reset timing, token volume, estimated cost, and activity history.

  • clipper: take a screenshot of any macOS window in one step from the terminal. pick with fzf, copy or save the image, and use --json when scripts or agents need the result.

  • winpick: switch to a macOS window from the terminal. pick with fzf, focus by window ID, list windows, and use JSON output for scripts or agents.

  • poplink: open files mentioned in a tmux pane without manually copying paths. It finds file paths, locations, stack traces, and markdown links, then opens the selected target.

  • transformer-from-scratch: English-German translation model built from scratch in PyTorch. Trained on WMT14 with multi-GPU training, beam search decoding, and reaching 25.53 BLEU on the test set.

  • scriptify: end-to-end handwriting synthesis system with PyTorch training, model packaging, FastAPI inference, and a React frontend.

  • tiny-adder-lab: tiny autoregressive transformer experiments for 10-digit addition using sub-300-parameter models.

  • pe-explorer: transformer positional encoding experiments across RoPE, sinusoidal, binary, decimal, and baseline variants for testing some of my hypotheses.

  • distributed-task-queue: go task queue with RabbitMQ, Postgres persistence, retries, idempotency keys, dead-letter handling, and Prometheus metrics.

Pinned Loading

  1. ai-deep-dive ai-deep-dive Public

    An open-source interactive learning platform for understanding LLMs through lessons, visualizations, and coding challenges.

    Python 7

  2. remux remux Public

    Native iOS tmux client with a mobile-first UI for persistent terminal sessions.

    Swift

  3. kbolt kbolt Public

    A local-first retrieval engine that turns notes, docs, and code into a searchable knowledge base for humans and AI agents.

    Rust 1

  4. scriptify scriptify Public

    A PyTorch handwriting-synthesis model built on Graves’ LSTM+attention paper, with multi-stage distributed training, a React demo UI, and a ready-to-use inference API.

    Python 13 4

  5. transformer-from-scratch transformer-from-scratch Public

    End-to-end, from-scratch PyTorch implementation of Transformer (Vaswani et al., 2017) with modular components, multi-GPU training, Weights & Biases experiment tracking, training scripts and ready t…

    Python 2

  6. hdfs-cloud-starter hdfs-cloud-starter Public

    Automate the deployment of Hadoop HDFS cluster on GCP & AWS using Terraform and Ansible.

    Shell