Julien Ziolkowski

Forward-Deployed Engineer · AI Automation

I turn slow, manual workflows into AI automations (agents, integrations and internal tools) that run self-hosted or on frontier APIs like Anthropic's.

01Profile

Forward-deployed engineer with 4 years at a Shopify Partner agency, taking AI systems for e-commerce clients from the first discovery call to production. I write the Go backends, MCP servers, prompts and data pipelines myself, run models self-hosted when the data is sensitive, and explain the results to people who don’t care how it works. On the side, I build ML and UI libraries from scratch in Odin.

02Experience

AI & Automation Engineer

beyondesign · Shopify Partner agency · CRO, A/B testing & performance marketing for D2C brands

July 2022 – present

Joined in May 2022 as Communication Associate (client comms, discovery calls). Formerly titled Communications Engineer.

  • Built and shipped the team’s in-house analysis platform in Go (~15k LOC, single binary, DuckDB, LLM orchestration). Per-client review and shop analysis went from ~30 hours to ~10.
  • Owned 30+ review-analysis and 10 data-analysis engagements from technical discovery and onboarding through delivery, presenting the results to clients directly.
  • Broke manual analysis workflows into steps a model can handle reliably, then wrote and iterated the prompts for each step against real client data.
  • Wrote custom MCP servers in Go that connect frontier models (mainly Claude) to proprietary client data.
  • Routed workloads between self-hosted models (llama.cpp, vLLM) and Claude by cost and data sensitivity, keeping PII-heavy data on our own infrastructure.
  • Turned vague client goals into measurable e-commerce KPIs and built the systems to track them.
  • Ran the team’s agentic workflows on a self-hosted VPS behind WireGuard for secure remote access.
  • Wrote the specs and architecture for an internal tool stack and design system built for agent-assisted development; coding agents implemented against them.

03Skills

AI & Agent Orchestration

  • Agentic workflows on hardened VPS
  • Routing local (llama.cpp) and frontier models by cost and data sensitivity
  • MCP servers in Go
  • Prompt design and iteration against real client data
  • ML fundamentals (PyTorch, tinygrad, own autograd library)
  • Building alongside coding agents against my own specs

Sovereign Infrastructure

  • WireGuard-secured private environments
  • Self-hosted, privacy-first deployments
  • Data sovereignty when PII or GDPR are in play
  • Single-binary Go deploys

Full-Stack Engineering

  • Go backends and frontends
  • Reusable, design-system-driven UI components
  • Internal tools and dashboards
  • Deterministic data pipelines
  • API and data integrations

Business Strategy

  • Technical discovery and client onboarding
  • Turning vague goals into measurable KPIs
  • Scoping and running client delivery
  • Explaining the technical side to non-technical stakeholders

04Projects

odin-ml

Odin · CPU & Metal

  • ML library from scratch, modelled on tinygrad. Lazy tensor graph, fused SIMD kernels, autograd on the IR, tested against tinygrad and MLX.

odin-ui-v2

Odin · Clay · raylib

  • Native UI stack with a widget library, force-directed graph view and a live design editor. Ships as a single binary.

dt-viz

Odin

  • Visualizations of Distribution Transformer models trained with odin-ml and drawn with odin-ui-v2.