Start from the workflow, not the model
Sit with the people who own the process. Agree on the problem, the data, the constraints and what a good outcome looks like. Write the spec and the architecture decisions before the first line of code.
02How I work
Forward Deployed Engineer: I work inside the customer's team, understanding their needs from the first definition of the use case to the system running in production. AI agents write a lot of the code; I own the scoping, the architecture, the review and what ships.
- discover
- scope
- architecture
- build with agents
- evals · tests · security
- deploy
- measure
- hand over
Small changes, measured behaviour
Prototype fast, then harden: retrieval, guardrails, permissions, fallbacks and hand-off to a human. Every change is reviewed and checked against evals, tests, security and cost before it ships.
Production from day one
CI/CD with a preview per change, monitoring and usage metrics from the first release. Then playbooks and building blocks the team can keep running without me.
04Stack
$ cat stack.txt
LLM agentsRAGEvalsOpenAI · Anthropic APIsClaude CodeTypeScriptPythonNode.jsReact · Next.jsReact NativePostgres · FirebaseAWS · VercelCI/CDObservability
// open to forward deployed engineering roles · madrid or remote · es/en
Let's talk about what you're deploying.
The assistant runs on papaia, a product I built. Prefer email? soygus@gmail.com