01LOCAL-FIRST AI LABORATORY
SHEYTAN Local Agent
A desktop AI engineering laboratory: managed llama.cpp inference, a real agent loop, isolated coding workspaces, and objective verification — all running locally.
GOWAILSLLAMA.CPP
RED SYSTEMLOADING
PARSA TAK — RESEARCHER · BUILDER · PROGRAMMER · WRITER · ARTIST
I research intelligence, build reasoning frameworks and local AI agents, engineer software, and turn the work into research, writing, experiments, and art.
CAPABILITIES
Capabilities backed by shipped systems — not aspirations. Each one is demonstrated by public work you can inspect.
Designing and building AI systems end to end: agent loops, controlled tools, context engineering, and verification gates.
Structuring how systems think: decomposition, critique, adversarial checking, and measurable benchmarks like AIST and ASI-100.
Shipping real software in Go, TypeScript, and React — architecture, tests, regression suites, and maintainable runtimes.
Running intelligence locally: managed llama.cpp inference, supervised agent loops, memory, and isolated workspaces.
Designing systems that hold: process supervision, state, bounded restarts, and deliberate evolution paths.
Turning research into frameworks: AI Instructions, REP, and USEF — governance, reasoning, and system improvement.
Engineering the web: static-first Next.js, performance budgets, accessibility, and honest, crawlable SEO.
Making technology expressive: canvas organisms, living interfaces, simulation, and generative art.
FEATURED WORK
Real projects with real destinations — every repository, article, and experiment below exists and is publicly reachable.
A desktop AI engineering laboratory: managed llama.cpp inference, a real agent loop, isolated coding workspaces, and objective verification — all running locally.
The measurement programme behind this laboratory: the AIST-2026.09 standard and the ASI-100-Elite benchmark, built on verified operational intelligence.
A free, open-source VPN configuration manager for Windows: a Go multi-core runtime that discovers, tests, maintains, and runs public proxy configurations.
A living canvas organism that turns this website into a computational experiment — adaptation, perception, and visible state on every page.
The site you are reading: a statically exported Next.js application that behaves like a living system — six scenes, a markdown-driven blog, and a generated SEO graph.
METHOD
One disciplined loop, applied to research, software, and art — understand before building, verify before shipping, evaluate after delivery.
Read the problem, the constraints, and the real goal before anything else.
Gather the material: sources, prior work, and the actual current state.
Judge what is true, what is missing, and where the real risks are.
Define the smallest change that produces a verifiable result.
Implement it as one coherent, reviewable step.
Test the result against reality: build, run, inspect, measure.
Fold what was learned back into the system and its documentation.
Ship it clean: static build, verified export, documented state.
Check the outcome after delivery and feed the next cycle.
WHAT I BUILD
I turn research into frameworks, software, experiments, simulations, and other working systems.
DIRECTION
The work is an ongoing attempt to understand intelligence, strengthen reasoning, and turn ideas into systems that can be tested and improved.
Explore intelligence, reasoning, and systems.
Turn research into software, frameworks, simulations, and art.
Build systems that are measurable, testable, and improvable.
KEEP EXPLORING
The website is a living index of the systems, experiments, writing, and art that come out of that process.
CONNECT
Public thoughts, updates, and signals.
Public channel for updates and broadcasts.
Visual work, experiments, and creative output.
Professional profile and network.
Videos, ideas, experiments, and published work.
Support the independent work and research.