AYRON JINS / LAB
SYSTEM: ONLINE
← HOME04 / PROJECTS
Layer / Projects & About

Ayron Jins projects.
Architecture is useful.
Execution proves it.

A working record of what I've built, tested and occasionally broken while figuring things out.

PROJECTS / SELECTED SYSTEMSEVIDENCE / CASE STUDIESSTATUS / ONGOING
05 / 05
Flagship project

The strongest work starts with a system problem.

AI COMPUTER AGENT / EXPERIMENTAL

From language model to computer-operating agent.

My flagship AI project looks at what happens once a language model stops just generating text. Give it controlled tools, put it in a real computer environment, and let it see the results of its own actions, and it starts acting less like a chatbot and more like part of a system.

USERLLMAGENT CONTROLLERTOOLSCOMPUTER / APPSOBSERVATIONLLM
KEY AREASTool calling · multi-step workflows · state observation · controlled interaction · permission boundaries · error handling.
WHY IT MATTERSIt demonstrates how an LLM becomes part of a larger software system rather than remaining only a text generator.
View full AI architecture ↘
Major systems

Projects that connect the layers.

INFRASTRUCTURE / ACTIVE

Self-hosted Home Lab

This isn't one application. It's an environment built for learning virtualization, Linux, containers, networking, reverse proxies, DNS, AI hosting, storage, media services and troubleshooting.

PROXMOXLINUXDOCKERPORTAINERSERVICES
WHAT I ACTUALLY DID
Deployed services, configured the environment, connected components, debugged failures and built supporting interfaces.
TECHNOLOGIES
Proxmox VE · Ubuntu Server · Docker · Portainer · AdGuard Home · Nginx · Immich · Nextcloud · Ollama · OpenWebUI · PaperMC.
LEARNED
Systems become understandable when the boundaries between host, guest, container, network and service are visible.
NETWORKING / ACTIVE

Self-hosted Network Infrastructure

A practical network path that keeps services reachable and easy to diagnose, without dressing it up as enterprise infrastructure.

DOMAINDNSROUTERNATPROXYSERVICE
PROBLEM
A service works locally but is unreachable from the internet.
APPROACH
Test application, localhost, listening ports, firewall, DNS, proxy, NAT and upstream path layer by layer.
LESSON
Debug the network path instead of changing settings randomly.
AI / ONGOING

Local AI Lab

An environment for experimenting with local language models, interfaces, retrieval, tool use and agent concepts.

LOCAL COMPUTEOLLAMALOCAL MODELOPENWEBUIEXPERIMENTS
WHY I BUILT IT
To understand model deployment and AI infrastructure rather than only consume hosted interfaces.
FOCUS
Model selection, prompts, RAG concepts, tool use, agent interfaces and inference infrastructure.
BOUNDARY
Local and interface components are described without claiming one universal runtime path.
DEVELOPMENT / EXPERIMENTAL

Dashboards, services and tools

Small systems built to make infrastructure easier to see and easier to run.

CODE SNAPSHOT
tool: create_document
permission: write
requires_confirmation: false
result: verify_output()
WHAT I ACTUALLY DID
Built interfaces, automation paths, service integrations and diagnostic views where the project required them.
STATUS
Experimental / ongoing; evidence and documentation are added as systems mature.
Project explorer

Inspect the work by system layer.

AI / EXPERIMENTAL

AI Computer Agent

A tool-enabled model architecture for controlled computer interaction and multi-step digital work.

Open related architecture ↘
PROBLEMHow can a model participate in useful computer tasks rather than only produce text?
ARCHITECTURELLM → controller → tools → computer → observation.
WHAT I LEARNEDPermissions, state, verification, failure handling and feedback loops matter.
Things that broke

A failure is only useful once you can isolate it.

PORT CONFLICTSFind the process or container that owns the required port.
DEPENDENCY ISSUESTrace runtime and package assumptions.
NETWORK FAILURESeparate local success from public reachability.
PROXY MISCONFIGURATIONFollow the request to its intended application.
MODEL / TOOL FAILUREInspect the result and choose whether to retry or stop.
How I approach projects

Build, break, understand, improve.

01UNDERSTAND
02DESIGN
03BUILD
04TEST
05BREAK
06DEBUG
07IMPROVE
08DOCUMENT

Define the problem, decide how components interact, create the smallest working system, verify assumptions, find edge cases, trace failures and document what changed.

Continue through the system

About the builder

A look at the interests and principles that shape how I build.

Open section →
Practical utility / Web architecture

Read what the crawler sees.

Use Site Intelligence Lab to inspect HTTP metadata, canonicals, robots.txt, sitemaps and structured data.

Open Site Intelligence Lab →