AI Process Engineer – Zürich.

Convolving

Zürich

Hybrid

CHF 140,000 - 200,000

Full time

3 days ago

Be an early applicant

A complete application in a minute

Tailored resume and cover letter, ready to send.

Get past ATS filters

Benefits offered by this job

Training budget

Job summary

Convolving GmbH in Zürich seeks an experienced software engineer to ship AI workflows into client environments. You will build retrieval pipelines, evals, and agentic systems across Claude, OpenAI, and Gemini, deploying on Azure, AWS, or GCP or on‑prem.

You will own the artifact end‑to‑end and work closely with a senior engineer during the sprint. The role requires 5+ years of software engineering with production LLM experience, strong Python and cloud skills, and comfort with client-facing

Qualifications

5+ years software engineering; 12+ months shipping LLM systems to production.
Fluent in at least two APIs: Claude/OpenAI/Gemini; strong opinions on retrieval and evals.
Comfort across one major cloud (Azure/AWS/GCP) including identity, networking, and data-residency primitives.
Python strong; TypeScript or Go a plus; SQL and data modelling.
Direct client-facing experience or indicators such as open-source projects, talks, or mentoring.
Allergic to abstraction-for-its-own-sake; ships working systems, not essays.
Swiss work permit or EU/EFTA citizenship; based in Oberrieden, ZH; 1–2 days remote per week.

Responsibilities

Build retrieval pipelines, evals, and agentic systems inside the client’s production environment across AI APIs.
Stand up evaluation suites using real workflow data; treat eval as progress unit.
Wire redesigned workflow into client stacks (ERP, CRM, document store, etc.) with proper auth and residency rules.
Own the controls and observability: trace model calls, logs, escalation paths for low confidence outputs.
Collaborate with Functional Expert to prototype against actual edge cases; hand over a maintainable artifact.

Skills

Python

TypeScript

SQL

Azure

AWS

GCP

Claude/OpenAI/Gemini

LLM systems

Client-facing

Job description

Full-time Zürich Posted 18 May 2026

01The firm

A senior firm, sized like a startup.

Convolving is a Swiss-based AI services firm running fixed-fee sprints for clients across five regulated industries. Delivery is partner-led on every engagement, and the team has shipped 30+ engagements to date. We pair tenured functional operators with AI-native engineers and a Technical Program Manager on every sprint, so the redesign lands in production rather than in a slide deck.

Engagements run 3 to 12 weeks at a flat fee. No retainer creep, no timesheet tracking, no change orders. When the artifact is shipped and adopted, the engagement closes. The hiring bar is senior across the firm: we do not build pyramids and we do not staff with juniors.

02The role

The engineer who ships into the client’s stack.

This is the engineering seat that puts the workflow into production. We hire AI Process Engineers out of two profiles: AI-native engineers from a research or applied-AI background who want client-facing work, and senior backend or platform engineers who have moved into shipping LLM systems in the last 18 to 36 months.

Your seat sits inside the client’s environment for the duration of the sprint. You ship retrieval pipelines, evals, agentic systems, and integration plumbing inside Azure, AWS, GCP, or on-prem footprints. The Functional Expert tells you where the controls are; the Technical Program Manager owns the cadence. You own the artifact.

First 90 days: pair with a senior engineer on two live sprints, take engineering lead on a third, and contribute to the internal eval and retrieval template library. By month six you are running one engagement at a time as the engineering owner, with a junior engineer or research apprentice paired in.

03What you’ll do

The work, week to week.

Build retrieval pipelines, evals, and agentic systems inside the client’s production environment across Claude, OpenAI, and Gemini APIs.
Stand up evaluation suites against the client’s real workflow data, not toy datasets, and treat the eval as the unit of progress.
Wire the redesigned workflow into the client’s existing stack (ERP, CRM, document store, ticketing, EHR, or function-specific systems) using their auth and data-residency posture.
Own the controls and observability layer: traceability of every model call, structured logs, escalation paths for hallucinated or low-confidence output.
Sit beside the Functional Expert in discovery and prototype against the actual exception cases they surface, not the happy path.
Hand over a maintainable artifact at sprint close: documented retrieval graph, eval suite, runbook, and on-call rotation handover to the client’s engineering team.
Feed every engagement’s patterns back into the firm’s internal template library.

04What we’re looking for

The shape of the hire.

Five or more years of software engineering experience, with at least 12 months shipping LLM systems against production traffic or a regulated workflow.
Working fluency with at least two of: Claude, OpenAI, Gemini APIs. Strong opinions about retrieval, evals, and structured output.
Comfort across one major cloud (Azure, AWS, or GCP) including identity, networking, and data-residency primitives.
Python strong; TypeScript or Go a plus. SQL and basic data-modelling fluency.
Direct client-facing experience or strong indicators that you can do it: presented at conferences, ran an open-source project with users, mentored a team.
Allergic to abstraction-for-its-own-sake. We ship working systems, not framework essays.
Swiss work permit or EU/EFTA citizenship. Based in our Oberrieden, ZH office; one or two days remote per week.

05Nice to have

Useful, not required.

Prior forward-deployed engineering experience at Palantir, Tenex, or a similar firm.
A published or open-source contribution to an evals, retrieval, or agentic framework.
Experience inside a regulated industry (finance, healthcare, legal, or pharma) including reading working knowledge of the controls vocabulary.
Reading knowledge of German, French, or Italian for European engagements.
Comfort writing customer-facing technical memos.

06What you’ll get

The package, plainly.

Comp. We pay senior market rates for the hub. Compensation is calibrated to Zürich market, not Swiss-discounted. Convolving GmbH on payroll. Base plus a sprint-completion bonus tied to adoption, not utilisation.

Sprint model. Engagements are fixed-fee, 3 to 12 weeks. No retainer creep. No timesheet tracking. You finish a sprint, you take a week before the next one starts.

Training budget. Annual CHF 5,000 for conferences, courses, or paid research time on a question that matters to your function or to the firm.

The work itself. One engagement at a time, with the room to push back on the design and own the artifact end-to-end.

Offerings