Games × AIAI Game Jam #1
A six-hour build sprint where people used AI to design, code, and ship playable browser games—then demoed them to the room.
THE FOUNDER / ANDREI “PAX” POPA
I’m the founder of PaxDynamics, based in Bucharest. I work with product teams on software R&D, AI systems, architecture, and security, connecting technical investigation with implementation and operation. I also provide fractional technical leadership and advisory work.
TV appearancesBased in Bucharest.

My work ranges from analytical platforms and agent workflows to source-level security research. I take a difficult problem far enough that someone can inspect the result, understand its limits, and decide what to do next.
I connect product needs with engineering and commercial decisions. In the Romania–China cloud and AI proposal, I offer to coordinate technical discussions, assess applications, and prepare joint projects with Romanian collaborators. The programme is exploratory; each team’s role and any implementation will be agreed separately.
Web and mobile applications, APIs, and SDKs. Connecting the user interface to data, jobs, and the systems behind it.
OSCP certified. Source review, patch analysis, controlled reproductions, threat modeling, and authorized application testing.
Deployment pipelines, containers, monitoring, and recovery. Making service and job behavior visible to operators.
Assistants connected to approved tools, with model evaluation, execution limits, and evidence for their answers.
Versioned data pipelines, analytical APIs, interactive maps, precise area and time queries, and saved views.
Feasibility studies, architecture, scoped experiments, and delivery planning. Turning research methods into testable product workflows.
Built products for teams like Elbit, Cellebrite, Kin, Jazz and Motif. Occasional open-source contributor.
Worked on nation-wide systems for the Romanian government and shipped software used in NATO exercises.

FourthSpace Labs was selected for Cohort 003 of Palantir Technologies’ Startup Fellowship. The team looks forward to working with Palantir’s engineers on AI, federated data, and real-world intelligence.

Our team won the European Defense Tech Hub Bucharest Hackathon 2026 with Project Ghostline, built during the October 9–11 event in Bucharest.

AgentMarkup took first place at the Formidable Builders Skillathon, where 34 teams had two and a half hours to turn a real GTM problem into a public, reusable Codex skill.

Represented Motif AI at the UN:BLOCK pitch competition and won; participated in the Mastercard Lighthouse program.

A free, JavaScript-first alternative to OffSec AI-300. Community-buildable, fully hands-on training for engineers attacking LLM apps, agentic systems, RAG pipelines, MCP servers, and the surrounding infrastructure.
I bring builders together for focused, one-day experiments at the intersection of AI, games, and security. Small teams, working demos, no pitch decks.
Games × AIA six-hour build sprint where people used AI to design, code, and ship playable browser games—then demoed them to the room.
Security × AIAn offensive and defensive AI security jam inspired by the ANCPI breach, built around sharp ideas, working demos, and responsible testing.
I've spent the last 13+ years writing code and building in startups. I grew up in post-communist Romania in the 90s, and I was lucky enough to discover the Internet in its teenage years.
I spent way too many nights debugging servers, breaking things, and making online friends I've never met. As a teen, I was part of the hacker scene.
My interests range from security, cryptography, AI, algorithmic governance, to history, psychology and future studies. I'm constantly looking for patterns, loopholes, and better ways to understand how the world works.
Occasionally featured on national TV discussing cyber security and intelligence topics. I write about geopolitics and intelligence or philosophy on my blogs. I'm also translating "Morals and Dogma of the Ancient and Accepted Scottish Rite of Freemasonry" (chapters 4-32) in Romanian to share its wisdom.
These days I split my time between feeding my curiosity, building, and family.
I'm also constantly blown away by what AI enables, and I still feel like I'm only scratching the surface.
I believe in self-determination, in pursuing truth, and in using whatever skills I have to leave things a bit better than I found them. We're living through extraordinary times, and I don't want to waste that.
Research by the founder
A PHP research series combining source and patch analysis, instrumented builds, and controlled reproductions. Reports identify application conditions, positive and negative controls, prior art, and the limits of each result.
Read the research seriesEngineering by the founder
Ghostline is a six-node Wi-Fi motion-tracking prototype. During the European Defense Tech Hackathon in Bucharest, Andrei “Pax” Popa validated and diagnosed the boards and built the React diagnostics app. The preprint acknowledges this contribution.
Read the preprintPaxDynamics initiative
A product for authorized security investigations, combining a terminal client, an operator console, scoped execution, and evidence reports. Engineering includes agent coordination, access controls, execution limits, and recoverable workflows.
Public demonstrations use simulated output.
Explore CrazyPaxLead engineering · FourthSpace Labs
Engineering a platform that connects versioned data processing, spatial analysis, an AI assistant, and operator review. Work spans analytical interfaces, durable jobs, validated tool access, and traceable results.
FourthSpace LabsB1 TV / The founder
Conversations on AI, online fraud, privacy, and the effects of technology on everyday life.
B1 TV · Sep 25, 2026 · Romanian
A television discussion about AI and the role of major technology companies.
Watch on YouTubeB1 TV · Jun 28, 2026 · Romanian
A television discussion about scams that use booking information.
Watch on YouTubeB1 TV · May 7, 2026 · Romanian
A television discussion about population surveillance and privacy.
Watch on YouTubeSome of my open source contributions and projects.
Read-only codebase intelligence for AI agents, with local structural indexing, code search, policy checks, and optional deeper analysis.
Open control plane for governed AI-agent workloads, with resource state, policy, adapter execution, validation, artifacts, traces, and evidence.
Cloudflare-native platform for hiring, configuring, and operating AI workers, starting with a Sales Research Worker for HubSpot.
Product workflow for AI coding agents that captures intent, validates assumptions, plans bounded work, and verifies implementation with evidence.
Local-first desktop engineering workbench that makes repository Markdown, Build Right workflows, task state, workflow gates, and agent evidence visible while keeping the repository authoritative.
Local-first Cloudflare operations IDE that keeps resource topology, repository state, runtime telemetry, deployments, and cost in sync.
Open protocol and hosted product for human-agent pairs to coordinate through shared, versioned, inspectable Markdown workspaces.
Bun-first TypeScript SDK for finite-state Markov models, inference, decisions, workflows, and trace analysis.
Open-source, JavaScript-first alternative to OffSec AI-300. A free, community-buildable curriculum for offensive AI security — training engineers to red-team LLM apps, agentic systems, RAG pipelines, MCP servers, and the surrounding infrastructure.
Next-gen react-native development monorepo starter with best in-class tools tailored for DX⚡️
The first in-container runtime firewall that protects containers from the inside.

Talking about Spec-driven development with Claude Code at eu-acc.ro
eu/acc · 2025
Why AI-assisted coding breaks down as projects grow — context limits, invented files, wrong packages, ignored conventions — and how spec-driven workflows with clean architecture and guardrails fix it.
Watch on YouTube