anton@portfolio ~
$ cat profile.json
{
  "name": "Anton Eliasson",
  "role": "Strategist & Agent Builder",
  "location": "Singapore",
  "building": "Kraken — multi-agent orchestration system",
  "background": ["Co-founder (9yr)", "CSO", "Enterprise partnerships"]
}

Co-founded Shakr (ad-tech, acquired 2024). Now building AI agent systems as Director of AI & Automation at VMG Digital.
Swedish · English & Swedish native, Korean conversational · APAC, NA, EMEA experience.

scroll
// about

Background

The short version

Swedish, based in Singapore. Co-founded Shakr — an ad-tech platform that generated video ads at scale for the world's largest brands and agencies. Grew it over 9 years as CSO, driving enterprise partnerships with Meta, TikTok, and global agencies. Shakr was acquired in 2024 by Shuttlerock.

What I do now

Director of AI & Automation at VMG Digital, serving 40-50 people across multiple teams. Built "Kraken" — a multi-agent orchestration system (1 orchestrator + 11 specialized sub-agents) — and established AI pods across 4 departments. Now commercializing internal AI tools into external products.

How I think

At VMG, strategists kept saying the briefs were bad. The obvious fix is a better template. But the real issue was that nobody had defined what a brief needs to become before it's actually useful — so I broke that down into 11 steps and built agents for each one. At Shakr, same instinct: we needed revenue, so instead of hiring more salespeople I got Meta to recommend us to their clients directly. I fixate on the problem behind the problem.

Enterprise DNA

Direct relationships with Meta (largest revenue driver at Shakr), closed a 7-figure licensing deal with TikTok, worked with Publicis, Ogilvy, Hogarth, Nike, Mercado Libre, Unilever, Nestlé, Alibaba, and Mercari. 20+ enterprise stakeholders across APAC, NA, and EMEA. 95% enterprise client retention rate.

// experience

Career

2025 — Present
Director of AI & Automation
VMG Digital · Singapore
Built VMG's AI function from the ground up, serving 40-50 people across multiple teams. Designed and deployed "Kraken" — a multi-agent orchestration system with 1 orchestrator + 11 specialized sub-agents, each powered by task-matched LLMs (Gemini 2.5 Flash, GPT-5.4, Grok 4.1 Fast). Also built an AI-powered LinkedIn ads analysis platform using multimodal LLMs. Established AI pods across 4 departments (7 people total). Now leading commercialization of internal AI tools into external-facing products.
Multi-agent orchestration LLM selection Product strategy Team building AI commercialization
2015 — 2024
Co-founder & CSO
Shakr → acquired by Shuttlerock (2024) · Seoul & Singapore
Co-founded and scaled a video ad technology platform enabling brands to create millions of personalized video ads at scale. Turned the Meta partnership into Shakr's largest revenue driver. Sourced and closed a 7-figure licensing deal with TikTok. Managed relationships with global agencies (Publicis, Ogilvy, Hogarth) and direct enterprise clients (Nike, Mercado Libre, Unilever, Nestlé, Mercari, Alibaba). Spearheaded transition to GPU-based rendering, reducing COGS by 10x. Scaled customer success across NA and APAC (13 direct reports), increasing revenue by 26%. Developed the Video-To-Go program with Meta across 20+ events in 3 regions, serving 600+ clients.
Enterprise partnerships Platform strategy Ad tech Global GTM Meta TikTok GPU rendering
2013 — 2015
Earlier roles
Seoul, South Korea
Marketing Associate — Spika Inc.2014–2015
Sales Development — EC212014
Digital Editor & Content — beSUCCESS2013
2011 — 2014
Education
Sweden & South Korea
University of Gothenburg — Communication & Media Studies (BA)2011–2014
Yonsei University — Intl. Management & Entrepreneurship2013
Gothenburg School of Economics — Business Admin2011–2012
Techstars Startup Mentor (2020–2022)
// kraken

What I Built: Kraken

Kraken is a production multi-agent creative brief orchestrator I designed and deployed at VMG. One orchestrator agent (GPT-5.4) coordinates 11 specialized sub-agents (all on Grok 4.1 Fast), with Gemini 2.5 Flash handling multimodal PDF ingestion. The system transforms raw client briefs into structured campaign strategy through three phases: foundation & audience analysis, strategy synthesis, and creative development.
USER Uploads PDF Brief INGESTION Gemini 2.5 Flash STORAGE Supabase ORCHESTRATOR Routes to agents, manages state, user conversation GPT-5.4 — never processes brief content ANALYSIS STRATEGY CREATIVE 01 Assumption Logger AUTO 02 Product Truth Report AUTO 03 Persona Mapper REVIEW 04 Platform Strategist AUTO 05 KPI Strategist REVIEW 06 4Ps Summary AUTO 07 FORCE Framework AUTO 08 The 5 Whys SELECT 09 Strategy Alignment AUTO 10 Main Ideas SELECT 11 Creative Narratives APPROVE All 11 sub-agents: Grok 4.1 Fast · 1.89s latency · 102 tok/s SUPABASE — SHARED PERSISTENCE Sub-agents write directly — bypassing orchestrator to prevent lossy summarization OUTPUT Structured campaign strategy
Analysis (autonomous)
Strategy (review/select)
Creative (user decides)
Gemini ingestion
Orchestrator (GPT-5.4)
DESIGN DECISION 01
Modular agents, not monolithic
A single model with all 11 frameworks in one system message couldn't maintain coherence — instructions bled across tasks. Isolating each step into its own agent with its own system message was the only way to get consistent output.
DESIGN DECISION 02
Database as output channel
Passing sub-agent outputs back through the orchestrator caused lossy summarization — the orchestrator would rewrite or condense details. Sub-agents now write directly to Supabase. The orchestrator only receives completion signals.
DESIGN DECISION 03
Session persistence through state
Progress stored in the database, not in memory. Users can leave mid-process, return days later, and resume at the correct step. Users can also go back and reprocess any previous step.
DESIGN DECISION 04
Human-in-the-loop at creative decision points
Analytical steps (1-7) run with minimal user input. Creative divergence steps (8, 10, 11) require explicit user selection before anything is committed. The system augments strategists, not replaces them.
Open full interactive presentation
// sierra_fit

Why Sierra

Agent design intuition

I designed and shipped a multi-agent system before applying to Sierra. Picked different models for different tasks, hit the architectural tradeoffs firsthand — where to isolate agents, how to handle state between them, when human review helps and when it's just friction. That means I can engage technically with Sierra's agent engineers during a build and with a prospect's technical team during discovery.

The founder-shaped strategist

The strategist role is about owning the success of a deployment: understanding the client's technology, identifying which agent use cases actually move the needle, scoping the POC, collaborating with sales, engineering, and product, and keeping the whole thing on track through to a long-term commitment. That maps closely to what I did at Shakr for 9 years. Different product, same shape of work.

9 years carrying deals

I spent almost a decade selling a technology product into enterprises where the buyer didn't fully understand what they were buying. TikTok, Meta, Publicis, Nike — every one of those required building enough trust that someone would bet budget on us. As a co-founder, those deals were personal. That changes how you approach the work.

A European in Asia for 12 years

I've lived and worked across Seoul and Singapore since 2013. Enterprise relationships with Mercari, Alibaba, agencies across the region. I'm not going to claim some special insight into every APAC market, but I've been here long enough that it's home, not an assignment.

// THESIS

The way I see it, the product Sierra is really selling is confidence. Every enterprise CXO considering AI for customer-facing roles is running the same calculation: what happens to our brand if this goes wrong? Sierra's job is to make that risk feel manageable. The tech matters, but the reason someone signs a multi-year deal is because they trust that Sierra won't embarrass them.

// presentation

Kraken — Full Presentation

vmg-architecture.html
Open in new tab →