Applied Agentic AI Engineer

We are building Applied AI directly into retail infrastructure. Not demos. Not wrappers around GPT. Real pipelines that process live inventory, enrich product catalogues at scale, and eliminate manual work that costs retailers hundreds of hours every month.

We are looking for an AI Engineer to own the core of this — the data enrichment engine that takes raw supplier files and outputs fully structured, enriched product catalogues ready for POS and online store, automatically.

What you will work on:

Building the AI pipeline that ingests messy supplier data (CSV, Excel, PDF, supplier URLs) and outputs clean, enriched product catalogues — descriptions, images, categories, variants, dimensions. Solving the image association problem at scale. Designing the model selection and batching strategy to keep inference costs sane at millions of SKUs. Owning the prompt engineering and fine-tuning decisions that make enrichment quality good enough that merchants never touch the output.

What we are looking for:

You have built AI systems that run in production, not just notebooks. You think about latency, cost, and failure modes, not just accuracy. You are comfortable owning the full pipeline from data ingestion to model output. You have strong opinions about when to use off-the-shelf models and when to build something custom.

Experience with LLMs, embeddings, and multimodal models matters more than a data science background. We do not need someone who optimises loss functions — we need someone who ships AI features that work reliably in the real world.

The stack: Python, cloud-native (Google Cloud), nShift, Stripe Connect. The problem space is retail, logistics, and product data at scale.

Why Vesko:

We are a small team moving fast with real customers, real transactions, and a clear roadmap to Series A. Business Finland funded. Google Cloud credits. You will not be the tenth engineer on a mature product — you will be the person who builds the AI layer from the ground up.

Based in Helsinki. Remote-friendly for the right person.

→ [email protected]