AI product engineer • LLM agents & devtools

Ship verifiable agent systems that stand up in production.

I design and build quality-dominant delivery loops for autonomous agents, developer platforms, and regulated workloads. The work is traceable: requirements → work items → code → tests → commits → telemetry.

Berlin or Remote Regulated AI workloads Developer productivity

Start a brief GitHub LinkedIn

Selected work

Highly opinionated projects, measurable impact.

Proof over hype. Each build leans on verification loops, instrumentation, and ruthless focus on outcomes.

Autonomous coding agent

Code Machine — Specs-to-code operating system

Code Machine is a general-purpose AI code generator that turns spec-based design artifacts into shippable software. I authored the empirical use case—requirements, success criteria, expected UX—and implemented the entire run, so the work spans product framing, research, and engineering.

Design artifacts arrive as structured specs (graph + requirements) encoding ownership, data flow, and dependencies between services.
LLM Code Ops orchestrates planning → building → updating for every node, grounding commits in the original graph.
Automated debugging loop runs the generated app, inspects logs, opens tickets, and feeds them back for self-correction.
Use-case narrative + evaluation written by me, ensuring the agent proves product value—not just code generation.
Flagship NL→DB proof delivered in 16.03 h for $71.66 with 156/156 unit+integration and 8/8 E2E passes.

View generated repo

PythonFastAPISQLAlchemyPyTestLLM Agents

Applied AI • Messaging

WhatsApp multimodal agent

Led discovery → MVP; launched a text+image assistant on WhatsApp with human handoff hooks and analytics-ready transcripts.

LLM OrchestrationVisionProduct Discovery

Rust • SaaS Gateway

dws-saas-layer

Rocket-based API gateway for authentication, billing, routing, and auditing across distributed services; Dockerized DX and hot reload.

View repo

RustRocketDockerSaaS

Conversational AI • Orchestration

ElGenie-Brian

Flask-based conversational brain handling stimuli routing, personality switching, cost calculation, memory, and OpenAPI3 endpoints.

View repo

PythonFlaskOpenAPILLM

Operating range

Systems + product + go-to-market literacy.

I move from architecture to delivery rituals without handing off context. Here’s how the stack breaks down.

Agents & evaluation

Designing orchestration, tool selection, and verification suites.

LLM AgentsTool OrchestrationPrompt EngineeringEvaluationGuardrails/Traceability

Platform & implementation

Shipping production services with durability + DX baked in.

PythonRustFastAPIFlaskSQL/SQLAlchemyDockerPyTest

Product & delivery

Clarity on the problem space plus quality gates before launch.

Discovery & FramingQuality GatesSystem DesignEnterprise AICI/CDREST APIs

Operating principles

Simple rules that keep agents honest.

The loop is verification-first, context-aware, and transparent for stakeholders.

Verification over vibes

Tests, linters, perf gates, and manual evals sit between every stage. Nothing ships without evidence.

Traceability by default

Requirements map to commits, PRs, and dashboards. The audit trail is a feature, not paperwork.

Context is a feature

Inputs are curated before any edit. Smaller context windows, higher fidelity outputs.

Human-in-the-loop ready

When autonomy stalls, there’s a fast manual override with the same telemetry in place.

Signals from shipping applied AI.

Long-form notes on building trustworthy AI systems, go-to-market loops, and monetization.

Substack • Systems

Solving the Last Mile Problem — From Agents to Systems

Field notes on making autonomous agents practical: how verification loops, traceability, and tight product framing close the “last mile” gap.

Read on Substack

LinkedIn • Business Models

Intro to the Open-Core Monetization Model

Breaks down how open-core plays support AI platform revenue—covering packaging, roadmap intent, and guardrails for sustainable growth.

Read on LinkedIn

Contact

Open to senior IC/EM roles and high-impact project briefs.

Berlin-based, relocation-ready, and available for new projects with teams shipping AI for regulated or high-stakes environments.

Let’s build it

I’m exploring roles in AI developer productivity and applied AI for regulated workloads, and I take on select project engagements. If that’s your team, let’s chat.

Email LinkedIn X

Quick links

Thesis repo — Code Machine

Request the resume (opens email draft)