Projects

⚊

ngx_markdown_filter_module

▢

An nginx filter module that converts markdown files to HTML on the fly. It uses the cmark library (or cmark-gfm for GitHub Flavored Markdown support) to parse and render markdown directly from nginx, without any static site generator or build step.

Features:

Real-time markdown to HTML conversion at the nginx level
GFM extensions: autolink, strikethrough, tasklist, tagfilter, tables
HTML template support with {{content}} placeholder
Works on proxy locations
Written in C, compiled as a dynamic nginx module

Fork of ukarim/ngx_markdown_filter_module with added GFM and table support.

📄 License: MIT 💻 Repo: github.com/bet0x/ngx_markdown_filter_module

⚊

barrahome-2026

▢

The source for this site. A markdown-only blog served by nginx using ngx_markdown_filter_module — no static generator, no build step, no framework. You write .md files, nginx serves them as HTML.

The stack:

nginx + ngx_markdown_filter_module for markdown rendering
A single bash script (barrahome_blog_gen.sh) to generate the post index
CSS styled as a retro CRT terminal with CDE Motif window decorations
Posts organized by date path: /YYYY/MM/DD/post.md

The entire publishing workflow is: write a .md file, run the index generator, commit, push.

💻 Repo: github.com/bet0x/barrahome-2026

⚊

vllm-router

▢

A fork of the vLLM project router — a Rust-based request router for vLLM inference deployments. The fork adds production and enterprise features that don't exist upstream.

Key features:

LMCache-aware routing — occupancy and prefix lookup modes, queries the LMCache controller for real KV cache state
PD disaggregation with independent prefill/decode policies and sticky session affinity
Routing explainability — x-vllm-router-* response headers showing worker, method, policy, cache status
Pre-routing hooks — HTTP callouts for safety, PII, and custom validation before routing
Model aliasing and fallback — rewrite model names, wildcard patterns, fallback chains
Decision export and replay — JSONL export of routing decisions, replay against different configs
Response caching — exact-match and semantic similarity with Redis backend
Semantic cluster routing — route requests by prompt content to specialized worker groups
OpenTelemetry tracing — opt-in OTLP distributed tracing with 7 stable span contracts
Admin API — config inspection, stats, decisions, graceful drain, hot config reload
Pre-built releases and Docker images on every tagged version

Written in Rust. Built on top of the upstream vLLM router foundation. Currently at v0.7.2.

📄 License: Apache-2.0 🐳 Docker: barrahome/vllm-router:latest 💻 Repo: github.com/bet0x/vllm-router

⚊

semlix

▢

A pure-Python full-text indexing and search library with semantic search capabilities, built on top of Whoosh. It combines traditional keyword matching (BM25/TF-IDF) with AI-powered semantic vector search, so a query like "authentication problems" can match documents containing "login issues" even without shared keywords.

Features:

Hybrid search merging lexical and semantic vector approaches
Multiple embedding providers: sentence-transformers, OpenAI, Cohere, HuggingFace
Flexible vector stores: NumPy and FAISS backends
Result fusion algorithms: RRF, Linear, DBSF
Fielded indexing, pluggable scoring (BM25F), spell-checking
Pure Python — no compilation required
Backward compatible with Whoosh

Available on PyPI: pip install semlix[semantic]

📄 License: BSD 2-Clause 📦 PyPI: pypi.org/project/semlix 📖 Docs: semlix.readthedocs.io 💻 Repo: github.com/semlix/semlix

⚊

broadcom_crawler

▢

A Python crawler that automates downloading Broadcom technical documentation and converts it to clean Markdown files with structured metadata. Designed to make vendor docs accessible for RAG pipelines and AI-augmented research.

Features:

Recursive crawling of Broadcom documentation pages
HTML to Markdown conversion with clean output
Hierarchical organization: docs/<product>/<version>/<section>/<page>.md
YAML frontmatter with source URL, product, version, and breadcrumbs
TOC extraction from dynamic JSON endpoints
Built-in rate limiting and preview mode
Kubernetes deployment configs included

📄 License: MIT 💻 Repo: github.com/bet0x/broadcom_crawler

⚊

readability-server

▢

A REST API server that wraps Mozilla Readability to extract clean, readable content from web pages. Send it a URL, get back the article content in HTML, Markdown, or plain text — with metadata like title, author, and publication date.

Features:

Multiple output formats: HTML, Markdown, plain text
Rate limiting, compression, and security headers for production use
Interactive API docs via Scalar and Swagger UI
Health checks and metrics endpoints
Optional API key authentication
Docker support with pre-built images on Docker Hub (barrahome/readability-server)

Built with Node.js, Express, and Mozilla Readability.

💻 Repo: github.com/bet0x/readability-server

⚊

docling-serve-sdk

▢

A Python SDK for the Docling Serve API. Type-safe document conversion with Pydantic models, async and sync support, and built-in connection pooling and retries.

Features:

Convert between 11+ document formats: PDF, DOCX, PPTX, HTML, images, and more
Multiple input sources: local files, HTTP URLs, S3
Configurable OCR engines and PDF backends
Hierarchical and hybrid document chunking
Table extraction
Async-first with httpx, full sync support as well
Custom error handling and retry logic

Available on PyPI: pip install docling-serve-sdk

📄 License: MIT 📦 PyPI: pypi.org/project/docling-serve-sdk 💻 Repo: github.com/bet0x/docling-serve-sdk

⚊

unsloth-docker

▢

A Docker environment for fine-tuning large language models using the Unsloth framework with GPU acceleration. Ships with Mistral Small 24B Instruct as the default model but works with any Hugging Face model.

Features:

Automatic Unsloth installation at startup
CUDA/GPU acceleration out of the box
Jupyter Notebook and JupyterLab interfaces for interactive training
Hugging Face model cache mounting to save bandwidth and startup time
Optional Flash Attention for enhanced performance
Multiple runtime modes: bash shell, Jupyter, or JupyterLab

Available on Docker Hub: barrahome/unsloth-docker

📄 License: Apache-2.0 💻 Repo: github.com/bet0x/unsloth-docker

⚊

openwebui-migrator

▢

A Python tool for migrating Open WebUI's SQLite database to PostgreSQL. Handles schema conversion, data type mapping, JSON fields, arrays, reserved keywords, and integrity checks — so you don't have to do it manually.

Features:

Complete schema migration from SQLite to PostgreSQL
Automatic data type conversion and mapping
Transaction-based safety with rollback on failure
Skips pre-populated tables to prevent duplication
Handles reserved SQL keywords and special characters
Comprehensive logging for debugging

Migrates all core Open WebUI tables: users, auth, chats, files, channels, and configuration.

📄 License: Apache-2.0 💻 Repo: github.com/bet0x/openwebui-migrator

⚊

manwrapper

▢

A simple Python command-line utility for viewing man pages in a straightforward, user-friendly way. No need to remember man command flags — just install and use.

Available on PyPI: pip install manwrapper

📄 License: MIT 📦 PyPI: pypi.org/project/manwrapper 💻 Repo: github.com/bet0x/manwrapper

⚊

bpaste

▢

A private, self-hosted pastebin service for the command line. Post, retrieve, update, and delete text snippets via HTTP with basic authentication. A simple alternative to public pastebins when you want to keep your data on your own server.

Features:

HTTP basic authentication for access control
Multiple posting methods: file upload, stdin, direct strings
Full CRUD operations on pastes
Shell script wrapper for easy CLI usage
Flat-file database storage (Lazer Database) — no SQL needed

Built with PHP and a shell wrapper. Inspired by cmdpb.

📄 License: GPL-3.0 💻 Repo: github.com/bet0x/bpaste

⚊

Hyper-V-Web-Console

▢

A web-based console for managing Microsoft Hyper-V virtual machines from the browser. Monitor and control VMs without needing RDP or the native Hyper-V Manager.

Features:

Display virtual machine information
Start, restart, and stop VMs
HTTP basic authentication
Web interface served from localhost

Built with Go and HTML/CSS/JS. Requires Windows with Hyper-V and PowerShell modules.

📄 License: GPL-3.0 💻 Repo: github.com/bet0x/Hyper-V-Web-Console

⚊

BashBlog

▢

A minimalist blogging platform built entirely in Bash as a CGI application. Drop a markdown file named YYYY-MM-DD_Title-of-the-post.md into a directory and it becomes a blog post automatically.

Features:

Automatic blog post generation from markdown files
CGI-based — works with nginx + fcgiwrap
Markdown parsing via John Gruber's original Perl tool
Simple installation with make install

📄 License: The Absurd License 💻 Repo: github.com/bet0x/BashBlog