ML
Mariana Lungu

Data architect · Northern Virginia

I build systems that can explain themselves.

I help regulated teams build data and AI systems that can be traced, defended, and trusted after the output leaves the screen.

See the problemRead the proof

Two houses · One source

One standard, expressed through systems and objects.

The technical practice

Archetype Core

Data and AI systems built for inspection: traceable, defensible, and ready to stand up after launch.

Enter the house

The problem I solve

Most systems look finished before they are defensible.

The output can look finished until someone asks where it came from, what changed, and whether it can still be trusted later.

01 · AI output

The answer looks clean.

But the team cannot prove which source, prompt, model state, retrieval path, or decision rule produced it.

When to bring me in

  • 01Govern the pilot before it becomes production.
  • 02Add lineage before leaders rely on the metrics.
  • 03Set the standard before speed becomes institutional risk.

The standard

I do not trust output I cannot trace.

Before I trust a system, I look for the record underneath it. Not a screenshot. Not a promise. A path someone else can inspect.

Question 01
Where did it come from?
Question 02
Can it be reproduced?
Question 03
What changed after it shipped?
Question 04
What happens under low confidence?

Provenance trace

Source to answer, with the record intact.

Each answer keeps its cited source, lineage, confidence decision, and audit record attached so another person can inspect the path later.

Confidence

0.91

Hash

SHA-256

Action

Review

The receipt

I built a retrieval system on federal policy documents to prove the standard instead of describing it. Every answer carries its evidence, decision context, and audit trail.

Tests

44

Indexed passages

9,116

Citations, confidence, model context, and source hashes stay with every answer.

Inspect the system

Writing

Public thinking as the work develops.

Newsletter

Beyond the Prompt

Video

YouTube

Still life of folded linen, ceramic vessel, and a sheet of paper in soft daylight

The origin

The decade that made the standard.

Federal systems, immigration cases, benefits processing, congressional inquiries: data that touched real people and had to be right. That is where my sense of provenance came from.

I learned to build where almost right was not good enough: where handoffs mattered, records had to survive scrutiny, and the quiet work underneath the visible system decided whether people could trust it.

This site is not my résumé. It is the public signal of that standard, and why I should be the person in the room when the system has to stand up later.

Read the story