Delight Labs

AI. Research. Investing. Products.

Delight Labs uses modern AI to dramatically increase the amount of useful software and research that a small team can create. Ideas that used to need large teams, budgets and months of coordination can now be explored by one capable person working with powerful models. This page lists what that has produced so far.

AI-powered software and automationGUI and CLI tools for developers and power users, automation of complex workflows, and hosted versions of these.
Finance, retail and investingMarket research, data analysis, portfolio tooling and automated investment workflows.
ResearchDistributed computing, physics and AI: papers, technical notes, experiments and short reports.

Apps

Desktop apps for macOS and Windows, web apps, and tools.

Delight Commander: two file panes side by side

Delight Commander Open source

A two-pane file manager in the tradition of Norton Commander, with ideas taken from Finder, Chrome and Sublime Text: tabs in each pane, list and icon views, a preview of the selected file in the opposite pane, and zip, 7z and tar archives browsed like folders. Every action is on the keyboard.

macOS on Apple silicon (.dmg), Windows x64 (portable .exe). The builds are not signed by Apple or Microsoft, so the first launch shows a warning.
Buffers: tabbed scratch-pad editor

Buffers Open source

A tabbed scratch-pad editor inspired by Sublime Text, for drafting emails, messages and LLM prompts before pasting them elsewhere. Text lives in tabs instead of files and is restored exactly after a restart.

macOS on Apple silicon (.dmg), Windows x64 (portable .exe). The builds are not signed by Apple or Microsoft, so the first launch shows a warning.
Stocks: grid of ticker cards with intraday charts

Stocks Web app

A stock research dashboard: a watchlist of price charts, screeners for trading and long-term investing, an earnings calendar, per-stock backtests, and AI opinions on a stock from local LLM command-line tools.

Shopping: Explore view with YouTube, Instagram and web results for brands

Shopping Not yet released

One cart and one list of liked items across Amazon, eBay, Chrono24 and Dubizzle, with the delivery promise, landed cost, condition and seller of each item. An Explore view collects YouTube videos, Instagram posts, news and web pages about the brands and products you follow. It reads pages you could read yourself and never buys anything on its own.

AI usage dashboard: limits for Codex, Claude and Gemini

AI usage dashboard Live

Tracks the usage limits of Claude, Codex and Gemini in one place, compares usage with an even use of each window, projects when a limit runs out, and keeps a weekly history.

Multiplex concept mockup: sessions from several machines and providers, two chats side by side

Multiplex Concept

One place for all your coding-agent chats. A small server on each of your machines connects to the Claude Code, Codex and Gemini CLI sessions running there, and the Multiplex app (web, macOS, Windows, Android) shows every session from every machine and provider side by side. Read along, reply, and approve commands from anywhere.

The screenshot is a design mockup; the app is not built yet.

Research

All papers on GitHub

First page

PaxosLease in Relativistic Inertial Frames

2026 · distributed systems, physics

Under special relativity, "at most one holder at any time" has no frame-independent meaning. The note restates lease safety with light cones and adapts PaxosLease: its containment and quarantine rules each gain the relativistic Doppler factor.

First page

Monte Carlo Experiments of Network Effects in Randomized Controlled Trials

2023 · A/B testing on social networks

When treatment nodes are surrounded by control nodes on a social network, part of their boost leaks into control and the measured lift underestimates a full rollout. Simulations on Watts-Strogatz graphs separate seven effects and show how clustering the treatment group reduces the bias.

Orgsim: simulated company dashboard
Interactive

Orgsim

An explainable simulator of a large company: 19,608 people in five management layers, and eight groups of institutional settings that decide how the company measures, rewards and controls work. Vanity metrics, killed ideas, empire building and the best people leaving all follow from people responding to what they are paid for, and the simulator measures how much value each fix would recover.

Books

In progress. Title pages and contents so far.

Cover of Reading Modern AI
Contents, page 1 Contents, page 2
In progress

Reading Modern AI

The papers on the road to large language models, read in order.

Cover of Distributed and Concurrent Algorithms with Python
Contents, page 1 Contents, page 2
In progress

Distributed and Concurrent Algorithms with Python

Sixty years of distributed algorithms, from mutual exclusion and logical clocks to Paxos, leases, Byzantine agreement and distributed transactions, each with runnable Python code.

Cover of The Engine Lineage Black Book
Contents, page 1 Contents, page 2
In progress

The Engine Lineage Black Book

id Software's game engines from Wolfenstein to Doom 3, read from the source code.

Writing

Bytepawn

Long-form technical posts on distributed systems, data science, A/B testing, physics and AI, most of them with working code.

Papers

Academic-style papers from 2007 to today, from spatial indexing of astronomical databases and lattice gauge theory to Paxos and experimentation.

Past delight: Scalien

Budapest, 2009 to 2012

Scalien was a NoSQL database startup founded in Budapest in 2009. It built two replicated databases on top of Paxos, ran them in production for a handful of clients, and closed at the end of 2012 when it could not raise venture capital. Both databases are open source.

Keyspace

Scalien's first database, written in 2009: a key-value store with strong consistency, fault tolerance and high availability, replicated with Paxos. The PaxosLease algorithm was designed for its master leases.

ScalienDB

A scalable, replicated database built on Paxos, developed from 2010 to 2012 using the experience from Keyspace. The paper describes its design decisions, parts of the C++ codebase, and the lessons learned running it in production.

About

The offer letter that started it.

DELIGHT LABS
Dubai · Budapest
Adriatic coast

Hi Marton,

I’m delighted to offer you a position as Member of Technical Staff at our organization.

At Delight Labs, our mission is simple: use modern AI to dramatically increase the amount of useful software and research that a small but capable team can create.

We work across three main areas:

  • AI-powered software and automation — using AI models to rapidly build GUI and CLI tools for developers and power users, automate sophisticated workflows, and create SaaS/cloud version offerings from these.
  • Finance, retail, and investing — exploring applications of AI and automation in finance and retail, potentially including quantitative investing, market research, data analysis, portfolio tooling, and automated investment workflows.
  • Research — conducting basic and applied research at the intersection of Distributed Computing, Physics and Artificial Intelligence, publishing papers, technical notes, experiments, and short research reports.

Delight Labs is deliberately small. We believe that AI changes the economics of software development and research: ideas that previously required large teams, significant budgets, and months of coordination can increasingly be explored by a small but capable team working with powerful models.

Your role as Member of Technical Staff is therefore unusually broad.

You are expected to identify interesting problems, understand them deeply, build things, run experiments, write code, read papers, develop new ideas, discard bad ones quickly, and occasionally turn good ones into software or research worth publishing. There is no strict separation between researcher, engineer, product manager, designer, and writer. The person who has the idea is generally encouraged to build the first version.

You will have considerable freedom to choose what you work on, guided by three principles:

  1. It should be interesting.
  2. It should benefit both humanity or Delight Labs.
  3. It should produce something tangible.

Tangible output might mean a working tool, a useful automation, an open-source project, an experiment, a paper, a research note, an investment system, or simply a well-documented negative result.

We value curiosity, technical depth, speed, intellectual honesty, and the willingness to actually finish things.

Your compensation package consists primarily of complete ownership of the company, unlimited access to management, an extremely flexible working environment (we have offices in Dubai, Budapest, and along the Adriatic coast), and whatever upside Delight Labs eventually manages to create.

Let’s agree on a start date as soon as possible.

We are excited to have you on board.

Best,
Marton
Founder, Delight Labs