KIWI is not just a company, it's a revolution. Since 2019, we have been passionately committed to creating innovative solutions that improve people's lives, helping as many smokers as possible quit smoking. With our headquarters in Croatia and a rapidly expanding international presence, our goal is to become the reference brand in Europe and a global benchmark in the e-cigarette industry.

Our products combine cutting-edge design, innovative technology and a strong commitment to sustainability, proving that changing the world is possible, one step at a time.

KIWI is looking for a Mid-Senior AI Engineer to join its international team.

The role

We're looking for a Mid-Senior AI Engineer to help us build and maintain our internal AI stack — a greenfield, privacy-first platform you'll help shape from the ground up. Hands-on role with direct exposure to the tech lead: solid operational autonomy on day-to-day work, with architectural and stack decisions discussed, mentored, and approved before go-live. You'll work closely with whoever drives the technical roadmap, with a clear path to grow into wider ownership over time — you won't be working in isolation.

What you'll build (core)

- Italian-language audio/voice AI pipelines on on-prem infrastructure (privacy by design)
- Autonomous LLM agent orchestration with cost, turn, and human-in-the-loop guardrails
- Custom MCP servers and tooling: resumable LLM session orchestration, integration with internal tools
- Italian NLP: sentiment and topic models, multilingual embeddings, domain-specific document classification

You'll also touch

- AI integration with the company's ERP and e-commerce platforms




- OCR on domain-specific documents
- Operational PHP work on existing e-commerce codebases (debug, maintenance, small changes)

Core technologies

- Python 3.11+, async ecosystem, testing, structured logging
- Applied LLMs: Anthropic Claude (SDK and headless mode), Ollama, local runtimes on Apple Silicon, Whisper
- Agent tooling: building custom MCP servers, resumable session orchestration, integration with internal tools
- Italian NLP: spaCy, Italian-specific sentiment and topic models, multilingual embeddings
- On-prem deploy on macOS, secure remote access
- Agentic Git workflow: parallel worktrees, issue-driven, PRs with mandatory human review

Must Have

- 2–4 years in AI/ML engineering or Python backend, with hands-on LLM experience (production exposure a strong plus)
- At least one NLP/AI pipeline or LLM feature shipped beyond the prototype stage
- Familiarity with the MCP ecosystem: either you've built a custom MCP server, or you understand tool design, schema, and error-handling well enough to build one with guidance
- Comfortable with Anthropic/OpenAI APIs, including tool-use and basic cost management
- Working PHP knowledge (debug, maintenance, small developments on existing codebases)
- Italian native or C1; fluent technical English




- Collaborative approach: able to discuss and justify technical choices, open to reviews, mentoring, and scope revisions

Nice to Have

- You've experimented with local LLM runtimes (Ollama, MLX, vLLM) and understand their trade-offs
- Experience with ERP platforms (Odoo or similar), even just on the integration side
- Experience with Magento or PHP-based e-commerce platforms
- Exposure to autonomous agents with guardrails (cost cap, turn cap, human escalation)

Engagement & compensation

- Independent consulting engagement via Italian VAT (P.IVA) — KIWI is a Croatia-based company and this role is structured as a professional services engagement with an Italy-based specialist who already operates as an independent consultant
- Monthly flat fee of € 4,000–5,000 VAT included for the services defined in the project scope, subject to renegotiation if scope or duration changes
- Consulting engagement tied to project milestones and deliverables, renewable based on outcomes achieved
- The overall fee takes into account agreed periods of unavailability planned in advance
- Accommodation and logistics costs are reimbursed for the on-site project ramp-up period in Buje
- Dedicated budget for training and conferences to keep growing

Full onboarding and shadowing on-site at Buje, Croatia for the first 6 months, with company-paid accommodation to make this phase as smooth as possible. After onboarding, we move to a hybrid/remote-friendly model, with the exact arrangement agreed together based on how the role and the team have settled in.

#J-18808-Ljbffr

📌 Mid -Senior AI Engineer (Friuli-Venezia Giulia)
🏢 KIWI VAPOR
📍 Friuli-Venezia Giulia