Engineering Leader · Solution Architect · Agentic AI

Sepehr Samadi

I build and lead the engineering behind mission-critical platforms that can't go down, and I do it by directing AI agents, not just writing code by hand.

Founder-CTO turned solution architect. Twelve years, three continents, hundreds of thousands of daily users, and one live trading platform being rebuilt for 15 institutional clients right now, with zero downtime.

See the work Work with me CV ↗

Sepehr SamadiHelsinki, Finland

Available for select advisory & consulting engagements

31Engineers hiredBuilt an engineering org from three people to thirty-four

93%Agent accuracyAutonomous support agent, live in production

15 · 0Clients · downtimeInstitutional clients on one core, migrated with zero outages

6Talks / 18 moInternational conference keynotes on agentic AI

SEC // 01 / WHAT I DO

Three things I'm genuinely good at

Whether you're hiring a leader, integrating AI, or making an architecture bet, here's the short version of where I'll move the needle.

D-01 · For engineering-leadership roles

Engineering Leadership

I build and run engineering organizations: hiring, structure, standards, and direction, not just the code inside them.

Grew a company 3 → 34 and personally hired all 31 engineers
Led cross-functional teams of 9–14 across backend, frontend, UX and infra
Final technical authority; reports to executives and client C-level

Founding CTO · Lead · Architect

D-02 · For AI engineering & integration work

Agentic AI Engineering

I make non-deterministic models behave like production software, with guardrails, determinism, and accuracy you can measure.

MCP servers & agent orchestration with OpenAI Agents SDK, LangGraph, CrewAI
Shipped a support agent at 93% accuracy that cut response time 65%
Six international keynotes on the topic in eighteen months

MCP · Orchestration · Determinism

D-03 · For architecture & advisory engagements

Solution & Platform Architecture

I design multi-tenant platforms that serve many clients from one codebase, and migrate live systems without downtime.

Replacing a live trading platform for 15 institutional clients, zero outages
A configurability model that avoids both over-abstraction and per-client forks
Polyglot full-stack: React/Next, Vue/Nuxt, Java/.NET/Node, AWS · Azure · K8s

Multi-tenant · Zero-downtime · Cloud-native

SEC // 02 / SIGNATURE WORK

Systems I've built where the stakes were real

Not a résumé dump. Just the three projects that best show how I think about high-stakes architecture, AI, and real-time systems.

FINANCIAL PLATFORM ARCHITECTURE0user-facing outages during a live rebuild

Zero-downtime replacement of a live trading platform

Solution Architect & Delivery Lead · FA Solutions · 2026–Present

Rearchitecting a legacy trading platform that serves hundreds of thousands of daily users across 15 institutional clients, cut over module by module with no maintenance window. I designed the configurability model that lets one product serve every client without a fork per customer, and I own the architecture end to end while staying hands-on in every layer.

15Institutional clients, live

9Engineers directed

C-levelWeekly client stakeholders

ReactTypeScriptJavaAzure

AGENTIC AI IN PRODUCTION93%accuracy from an autonomous support agent

An AI agent that resolves support tickets on its own

Lead Engineer & Technical Product Owner · Metacore Games · 2024–2026

Designed and shipped an agentic system that reads a player ticket, works through the platform, and drafts a resolution: 65% faster responses at 93% accuracy. In the same role I led the ground-up rebuild of the internal admin platform (~130 users) to 100% adoption in three months and quadrupled deploy frequency. Recognized with the company-wide Act of the Year 2025.

65%Faster ticket response

4×Deploy frequency

100%Rebuild adoption in 3 months

ReactNext.js.NETAWSKubernetes

REAL-TIME EMBODIED SYSTEMSReal-timeperception-to-motor control loop

Brains for a humanoid robot deployed in airports

Senior Software Engineer · AiCell · 2020–2022

Built the interface for a humanoid robot deployed as an assistant in international airports, and contributed to the event-driven core behind it, reacting in real time to vision and language modules and driving motors and hardware. The physical origin of the problem I still solve today: building predictable systems around components whose output can't be predicted.

AirportsLive international deployment

3Modules fused: vision · language · motor

FlutterPythonEvent-driven core

SEC // 03 / THE MANAGEMENT UNDER THE TITLES

I've led people, not just code, in every role, whatever the title said

My titles have read Engineer, Lead, Architect and CTO. In each one I owned hiring, team shape and technical direction. Here's that record made explicit.

3 → 34Company built as founding CTO

31Engineers personally hired

9–14Typical team size led

21Projects delivered for 15 clients

Co-founder & CTO · Vorna

Built the whole org

Grew from 3 to 34, hired all 31 engineers, set the technical bar and ways of working, and stayed final authority across 21 concurrent projects. Revenue grew ~70% then 200% year over year.

Lead Engineer & acting PO · Metacore

Ran the team and the product

Led a team of twelve and acted as product owner for a platform used by ~130 people, owning definition, planning, tracking and stakeholder reporting, not just the code.

Solution Architect · FA Solutions

Sets direction, faces the executives

Leads nine engineers, sets the technical direction every client variant is built on, reports bi-weekly to internal executives and weekly to client C-level.

SEC // 04 / AGENTIC AI ENGINEERING

Making LLMs behave like production software

The thread running through six conference talks and every recent role: engineering deterministic guarantees around non-deterministic models.

AI-01

Determinism around non-deterministic models

The problem I've built a career around: guardrails and determinism design so an LLM feature behaves predictably in production, measured on accuracy against a target, not vibes.

AI-02

MCP servers & agent orchestration

Building and directing agent systems with MCP servers, the OpenAI Agents SDK, LangGraph and CrewAI: tool calling, multi-agent choreography, and the plumbing that lets agents act on real systems safely.

AI-03

AI-native delivery, end to end

I build by directing AI agents, not writing every line by hand. I built a pipeline that turns a bug report into a plan, dispatches a coding agent, opens a PR and hands it to a human, and drive that model across coding, review, docs and tickets.

AI-04

Cost, latency & token discipline

Treating LLM calls as a production resource with a budget: prompt and context design, model selection per task, and caching so agentic features stay fast and affordable at scale.

Model Context Protocol//OpenAI Agents SDK//LangGraph//CrewAI//93% production accuracy

SEC // 05 / ON THE RECORD

Recognized beyond my own teams

Keynotes, peer-reviewed research, mentorship and awards. The external signal that backs up the claims above.

6Conference keynotes in 18 months

3Peer-reviewed papers · 22 citations

Top 10AI Design Mentor on ADPList

Guest lecturer, Software Architecture, Aalto University
Act of the Year 2025, company-wide award, Metacore
Mentor, We in Games Finland

Selected conference keynotes

Architecting for Non-Determinism: System Design Patterns When Your Core Logic Is an LLM
BaselOne 2026 · Basel
Drupal as an MCP Server: Making Your CMS a First-Class Citizen in the Agentic Web
DrupalCon 2026 · Rotterdam
Lost in Translation: When EU AI Governance Meets the Codebase
AI Summit 2026 · Brussels
Designing Deterministic Systems Around Non-Deterministic LLMs
AgentCon 2026 · Rome & AgentCamp · Barcelona
When Your Pair Programmer Is a Swarm: Architecting UIs for Multi-Agent Coding
AI Native Devcon 2025 · New York

SEC // 06 / WORK WITH ME

Bring me in where the architecture bet is expensive to get wrong

Alongside my role I take on a small number of advisory and consulting engagements: the kind where an experienced second set of eyes changes the outcome.

AI & agentic integration

Where an LLM belongs in your product, how to bound its non-determinism, and what a realistic accuracy and cost target looks like, before you build.

Architecture review

A sharp, time-boxed second opinion on a system's architecture, multi-tenant model, or migration plan, before a costly decision is locked in.

Technical due diligence

An outside read on engineering health, team structure and architectural risk, for investors or leadership evaluating a codebase or a team.

Emailsamadisepehr@gmail.com

LinkedInlinkedin.com/in/sepehr-samadi

Based inHelsinki, Finland (UTC+2)