Dr. Eidan J. Rosado

I build AI systems people can understand and trust.

Research scientist. Engineer. Author.

I’m Eidan, the creator of Miilo. I build open-source tools for privacy, agent behavior, and human-centered AI, and write about what it takes to make them work.

Explore my work Meet Miilo

Research, creativity, and the space between.

Currently building

Clarity for
work in motion.

A quieter way to stay informed.

See what your agents are doing, across the computers you use. Follow the work, understand what changed, and notice when something needs you.

I built Miilo to make agent activity easier to understand without replacing the tools people already trust.

Open sourceListed in Omarchy & OmalauncherMobile in development

Explore Miilo Source code

Miilo’s orbit view showing an automation factory, coding sessions, and a resting system — Different tools. One place to see the work.

Ideas into working systems

Selected work

The questions behind my work: what should an agent do, what should stay private, and how do we understand what happens next?

Praxa

Releases

Make agent policies explicit, versioned, and easier to evaluate.

A research library that treats policy as a first-class object for language-model agents. Praxa represents policies as typed, versioned documents, then studies operationalization: how a stated policy becomes practice on a task through swappable methods such as base prompts, full-policy system prompts, retrieval, or few-shot exemplars. Keeping the policy fixed lets those methods be compared without rewriting the rules.

Explore a Python example

from praxa import Condition, Disposition, PolicyBuilder

policy = (
    PolicyBuilder("demo.medical.v1")
    .domain("medical")
    .rule(
        "MED-GEN-001",
        disposition=Disposition.ALLOW,
        conditions=[Condition(kind="tag", value="general_info")],
    )
    .build()
)
# same policy, many operationalizers to compare

View on GitHub

The Hedon Project

Robotics research

Explore how machines perceive, remember, and respond to experience.

Hedon is an open robotics initiative exploring affective, experience-aware machines. Its first platform, the UGV Beast, is a multimodal rover that senses through vision, audio, and optional biosignals, fuses those signals through a behavior engine, and acts on either a digital twin or live hardware. Experience logging feeds the device-agnostic memory research at the center of my current work, so what the rover learns is not locked to the rover.

View on GitHub

PII-Codex

Releases

Understand what sensitive information a system is handling.

An open-source Python library for detecting, categorizing, and assessing the severity of personally identifiable information in text. Published in the Journal of Open Source Software, it gives researchers and practitioners a shared taxonomy and evaluation framework for privacy-preserving analysis, and it underpins my later work on tool-augmented PII detection in small language models.

Explore a Python example

from pii_codex.services.analysis_service import PIIAnalysisService

PIIAnalysisService().analyze_item("Hi! Call Jane Doe at 555-0142.")
# PERSON, PHONE_NUMBER  →  Identifiable
# Hi! Call <REDACTED> at <REDACTED>.

View on GitHub JOSS Paper

The person behind the projects

An engineer’s habits.
A researcher’s curiosity.

Explore my research

I’m a research scientist, adjunct professor of computer science, and the founder of Viratec Interactive. My work brings together privacy engineering, agent behavior, and the human side of technology.

More than fifteen years in software and systems engineering have taken me from defense simulation to cloud platforms and AI systems. A Ph.D. in Information Systems and master’s degrees in Computer Science and Psychology give me different ways to ask the same question: how can we build systems that work better for people?

I teach mobile architecture, software testing, and responsible AI, serve on doctoral dissertation committees, and write the Cloud Bytes Collection. Away from the keyboard, I read, garden, and make art. A sketchbook and a well-designed system have more in common than you might think.

From practice to the page

Writing for people who build

Engineering Best Practices for the Agentic Developer

You've shipped something with Cursor, Claude Code, Replit, or another AI tool, and a small voice in the back of your head is whispering that you're not sure it's safe yet. You're not wrong. There is an entire discipline sitting between "it works on my machine" and "a stranger just trusted me with their data," and this book covers the parts that matter most when you're shipping software with the help of AI.

Inside: writing specs the AI can actually build against, setting up the AGENTS.md file that quietly doubles every coding session, keeping your secrets out of your repo and your cloud bills out of four digits, and writing the smallest tests that catch the biggest bugs. Each chapter closes with what to do Monday morning, a short war story of what happens when the lesson is ignored, and a prompt template you can paste straight into your AI tool.

View on Amazon All Books

A foundation in building real systems