Independent AI research · Est. 2020 · Manila → AmsterdamA transmission, in five chapters

We build AI systems. We write the words that make them human.

A lab where rigorous research and craft writing collide — published, shipped, and held to the same standard.

17yrs
In tech / writing
2025
Preprint · Tri-Persona AI
3 → 1
Agents · One truth
CH. I·The LabWhere

A small lab. Intentionally. Quality doesn't scale by headcount.

SG Research started with a question: why does most AI-generated content feel empty, and most technical writing feel inhuman? The answer is usually the same — whoever built it optimised for volume, not for truth.

We're led by Alexander V. Neri — technical writer, researcher, 17+ years across enterprise software, game development and machine learning. Founded in Manila. Headquartered in Amsterdam. Working with clients worldwide.

We treat content as a precision instrument and AI as a research subject, not a shortcut. When we build AI systems, we publish what we learn. When we write, we write like we mean it.

2025
Tri-Persona AI preprint published
A multi-agent debate system for OSINT/GEOINT analysis.
2023 — 24
Operations moved to Amsterdam
Expanded into European enterprise clients; ML research formalised.
2020
SG Research established
Founded in Manila, Philippines. Writing + AI from day one.
2008 — 20
Industry foundation
Enterprise software, game development, Microsoft Azure ecosystems.
CH. II·InquiryPreprint · 2025

Teaching machines to argue with themselves.

OSINT analysts are drowning in data. Single-model AI hallucinates, misses context, and delivers false confidence. We took a different approach.

Inspired by the MAGI supercomputer from Evangelion, our architecture deploys three distinct AI personas that actively debate conclusions before surfacing them. The result: explicit reasoning chains, reduced hallucination, and a built-in audit trail for every inference.

// 01 — The Optimist
Melchior
hypothesis_generate()
Generates hypotheses. Finds patterns. Explores possibility space.
↗ exp(possibility)
// 02 — The Skeptic
Balthazar
fact_check()
Challenges every claim. Scores source credibility. Finds the holes.
⌂ doubt(0.62)
// 03 — The Judge
Caspar
synthesize()
Synthesizes the debate. Assigns confidence. Produces the verdict.
∴ verdict(0.74)
magi-system · tri-persona-analysis · v0.9.1
Melchior[optimist]→ hypothesis_generate()
"Three satellite passes confirm anomalous thermal signature, sector 7."
Balthazar[skeptic]→ fact_check()
"Source credibility: 0.62 · Temporal gap: 14h — insufficient."
Caspar[judge]→ synthesize()
"Confidence: 0.74 · Flag for analyst review · Recommend 2 additional corroborations."
$ export report --format=intel
CH. III·PracticeWhat we do

Three crafts. One studio. We write, we strategise, we research.

Engagements run from a single deliverable to multi-quarter R&D. Whatever the shape, the standard is the same: we publish what we learn and we ship work we'd put our name on.

S—01

Creative & Technical Writing

Documentation engineers actually read. UX copy that converts. Narratives that make complex systems feel inevitable — from onboarding flows to game dialogue to research briefs.

S—02

AI Content Strategy

Not another chatbot. We design intelligent content pipelines — knowledge bases, auto-generation systems, editorial AI — that scale output without losing voice.

S—03

AI Research & Consulting

Bring us your hard problem. We work at the research level — multi-agent systems, data analysis, ML integration — and translate findings into decisions your team can act on.

CH. IV·EvidenceSelected work

The proof is in the receipts.

Four engagements that tell the shape of the studio: writing, localisation, AI strategy, research. Different briefs, same standard.

01
Technical writing
Enterprise onboarding revamp
Restructured a 100+ page onboarding manual for an enterprise SaaS platform. Redesigned information architecture, rewrote for self-paced learning, and introduced modular sections engineers could update without breaking the whole doc.
40%
Drop in onboarding tickets
02
Creative writing
Game localisation & narrative
Adapted in-game dialogue and item descriptions for a mobile RPG, localising a Japanese IP for English audiences. Preserved original tone while making it feel native — a balance most localisation projects get wrong.
4.8★
Player narrative rating
03
AI content strategy
Fintech knowledge assistant
Built a GPT-based chatbot trained on a fintech startup's documentation. Reduced support volume by giving users instant, accurate answers — and surfaced content gaps the team didn't know they had.
60%
Support query reduction
04
AI research
Tri-Persona MAGI system
R&D on a multi-agent debate architecture for OSINT analysis. Three AI personas (Optimist, Skeptic, Judge) argue their way to more reliable conclusions. Reduces hallucination; produces explicit reasoning chains.
Preprint · 2025
CH. V·SignalEnd of transmission

Send a signal.

If you have a hard writing problem, a research question, or an AI system that needs to feel human — we'd like to hear about it.

We typically respond within two working days. Engagements start with a 30-minute call — no slide deck, no pitch.

Or — send a longer brief
SG · ResearchAmsterdam · ManilaMulti-agent intelligenceWriting · AI · ResearchEst. 2020Tri-persona MAGI · 2025
Direction

Alexander V. Neri

Principal · SG Research

Disciplines

Writing

AI strategy

Multi-agent R&D

Coordinates

Amsterdam, NL

Manila, PH

52.37° N · 4.90° E

Channels

hi@sg-r.net

ORCID · 0000-0001-7743-9546

Privacy

© 2026 SG · Research — All rights reserved
Built in Three.js · GSAP · Lenis