Independent research & engineeringBerlin · Working internationallyPortfolio / 2026

Ivan Andrianov.

I / The practice

Public information,
made rigorous.
Research,
given form.

I build the systems between a source and a decision: research infrastructure, analytical tools, and software for the public interest.

Selected work
Ivan Andrianov
Researcher. Builder.Political research / OSINT / Software
01—26
From public records to working instruments.Curriculum vitae PDFScroll to the work

II / Selected work

A body of work.

Three systems.
Different questions. The same discipline.

01 / Research infrastructureExercitusMilitary exercises, with the evidence attached.Ongoing researchDetailsClose
01 / Research infrastructure Ongoing research

Exercitus

The record.
The source.
The space between.

A registry of military exercises in the Baltic Sea region. Every date, participant, and personnel figure stays connected to its evidence.

Record structure2024 / Baltic Sea region
Umbrella exercise01
Steadfast
Defender 2024
90,000declared personnel
NATO · Source-linked
linked_to
Related exercise02
DEFENDER 24
40,000declared personnel
U.S. Army Europe and Africa
Related exercises. Separate counts.
October 2026 snapshot.
1,658exercise editions
2015–2026period covered

As of October 2026.
Reported activity, with documented limits.

The question

How can reports on military exercises become records whose evidence can be inspected? A personnel count may describe a whole exercise, a national contingent, or a related programme. Those figures are not interchangeable.

The system

Language-model extraction, deterministic checks against source text, and human-signed corrections. Source passages stay alongside values; umbrella exercises and components retain explicit relationships and identifier histories.

Scope & limits

1,658 exercise editions, 2015–2026, drawing on 29 official websites as of October 2026. Coverage is uneven; this is reported activity, not a complete census or an escalation prediction system.

An earlier single-reviewer assessment found 85.5% of 379 judgeable registry values correct. A separate held-out extraction test found 91.3% of 126 values correct. These measure different things.

Explore the registry
02 / Civic technologyjobsplzA walkable route to work in Berlin.Public toolDetailsClose
02 / Civic technologyBerlin, Germany
A different kind of job search.
010203YOUR POSTCODENEARBY PLACES
Before you walk inKnow what
you’re owed.
Workers’ rights, included.
A walkable round, not a list of adverts.

jobsplz

The city is full
of places to start.

Enter a postcode. Get a walkable round of places that hire off the street — and the rights information you need once you find work.

A city database, a weekly verification pipeline, and an interface built around the person on foot.

The question

Where can you walk in and ask for work in Berlin — and what are you owed once you do? A postcode becomes a walkable round, rather than another list of adverts.

The system

A city database answers searches locally. A weekly pipeline refreshes places, while an audit rebuilds the city from a second source to expose disagreements. Built end-to-end, including workers’ rights.

Scope

40,488 places, 96 Kiez pages, four interface languages. Figures from September 2026, not live counters.

03 / OSINTThresholdAn archived system. A more precise question.Archived researchDetailsClose

OSINT / Research archive

Threshold

An earlier system for organizing public reporting. Its evaluation changed the question — and led to Exercitus.

Archived
with its findings

The system

A research system for collecting and classifying public reporting across regions and time. I built the ingestion, classification, database, API, and frontend.

What changed

Evaluations across Ukraine, the Baltic region and the South Caucasus found that the escalation index did not consistently outperform a free baseline using GDELT Goldstein scores. The index was withdrawn from publication.

What followed

Missing historical exercise data led to Exercitus: a narrower registry with explicit evidence and uncertainty. Its evaluation does not establish improved escalation prediction.

View the code
Research infrastructure / Public tools / Open inquiry

III / Approach

A question is
only the
beginning.

About my practice

The work is in making it
possible to answer.

My background is in international relations and North American studies. I build the pipelines, interfaces, and analytical frames that let a question meet the evidence.

01

Keep the source close.

A useful record should let you inspect where it came from, not just accept the result.

02

Make uncertainty legible.

Disagreement, missing information, and failed tests belong in the system.

03

Build something usable.

The interface is part of the research. A finding matters more when someone can work with it.

IV / Writing

From the workbench.

All writing

V / Correspondence

Based in Berlin. Open to collaboration.

A serious question
deserves a conversation.

Research engineering, OSINT, risk intelligence,
or a problem that needs its own instrument.

ivan@evandrianov.pro