MJ Hwang
MJ Hwang is an independent full-stack/backend engineer and open-source creator focused on reliable, local-first AI service plumbing: collection, extraction, structured search, RAG, fact-checking, agent safety, and LLM operations. His stated view is that useful software will come from better infrastructure around models rather than simply larger models.
At a Glance
- Developers and AI-agent builders
- Researchers and technical users operating local LLMs
- Government and enterprise teams, including air-gapped environments
- Data engineering, BI/OLAP, analytics, and legacy-database migration teams
- +1 more
AI Tools by MJ Hwang
(1)
Claude Code Hidden Problem Analysis
Claude Code Bug and Quota Research
Discussions
No discussions yet
Be the first to start a discussion about MJ Hwang
Latest News
I Tracked 42,363 Claude Code API Calls. Here's Where Your Quota Actually Goes.
Testing Claude Code Against Local 35B Models: Building a Cross-Check Harness
I Built a 3-Node Home LLM Lab. Here's What It Actually Takes.
Anthropic's Postmortem Told Half the Truth
Products & Services
Open research repository analyzing Claude Code client-side bugs, cache behavior, rate-limit headers, quota/token inflation, model behavior, and release/changelog evidence. Includes proxy data, cross-validation, bug tracking, advisory material, and a quick-start guide.
Personal data-asset and LLM-native tool ecosystem. It provides composable local tools for collecting, extracting, searching, monitoring, guarding, and relaying AI-agent data flows.
Unified LLM usage-management layer: API proxy with full request/response capture, session diagnostics, multi-CLI orchestration for Claude Code/Codex/Gemini, and automated recovery.
Local autonomous AI agent with a custom ReAct pipeline, RAG using Qdrant and Neo4j, persistent memory, and a two-tier LLM fallback; designed to run on-premises without a framework or cloud dependency.
Market Position
Hwang positions his work as the infrastructure/plumbing layer around AI models: local-first, composable, MCP-native tools for collection, extraction, search, monitoring, safety, and LLM operations. This differentiates it from cloud-dependent AI-agent stacks and hosted observability/API platforms by emphasizing offline operation, direct measurement, transparent diagnostics, and small tools that can be chained together.
Leadership
Founders
MJ Hwang
Full-stack engineer with 6+ years building data-intensive systems; SQL-native tuning across PostgreSQL, Oracle, MariaDB, MSSQL, DB2, and Netezza; backend work in Python and Java; full-stack, mobile, data-warehouse/data-mart, BI/OLAP, analytics, and AI-service-pipeline experience. GitHub identifies him as ArkNill and lists Seoul, South Korea.
Executive Team
MJ Hwang
Independent full-stack engineer and creator (ArkNill)
Backend/full-stack engineer with 6+ years in data-intensive systems, SQL tuning across six database engines, Python/Java backend development, mobile/web applications, and end-to-end AI service pipelines.
Founding Story
The public materials describe a personal, project-driven practice rather than a separately incorporated company. Hwang built AI/data tools for his own pipelines—collecting news from hundreds of RSS feeds, extracting content for fact-checking, searching documents semantically, and monitoring pages—and extracted useful modules into standalone open-source packages. The LLM relay work began after his investigation of Claude Code cache and token-usage anomalies, leading him to build a transparent proxy and diagnostics layer.
Business Model
Revenue Model
The public materials describe open-source distribution rather than a commercial pricing model. QuartzUnit packages are published on PyPI under MIT or Apache-2.0 licensing and run locally; no paid subscription, API-usage, license-sales, or services revenue model is stated.
Pricing Tiers
QuartzUnit OSS packages are published on PyPI and described as open-source under MIT or Apache-2.0 licenses.
Target Markets
- Developers and AI-agent builders
- Researchers and technical users operating local LLMs
- Government and enterprise teams, including air-gapped environments
- Data engineering, BI/OLAP, analytics, and legacy-database migration teams
- Organizations needing local/on-premises AI pipelines and observability
- AI-agent data collection, extraction, semantic search, and monitoring
- Local and air-gapped enterprise or government data/AI pipelines
- Fact-checking and evidence-backed research
- RAG systems and autonomous agents with persistent memory
- LLM API observability, quota/token diagnostics, and multi-CLI operations
- Local LLM inference benchmarking and production configuration