AI Product Engineer

Understand Tech; Frontignan; On-site; EUR 60,000 - 100,000; Full time

Job summary

Understand Tech is expanding its AI platform for enterprises, delivering on-site AI factories and AI-in-a-Box solutions for customers in France and abroad. You will own your scope from roadmap to support, direct collaboration with demanding enterprise clients, and you will iterate quickly in a hands-on, growth-oriented environment.

You will build AI components, optimize LLM inference on NVIDIA GPUs, and operate models in production on customer hardware, with a focus on performance, governance

Qualifications

• Proficient in Python and modern Python web frameworks.
• Experience building end-to-end RAG pipelines: embeddings, vector stores, evaluation.
• Experience serving LLMs on NVIDIA GPUs with optimization techniques.
• Knowledge of PostgreSQL and at least one NoSQL DB (MongoDB, Firebase).
• Familiarity with Docker, Linux, CUDA drivers; basic cloud exposure (AWS/GCP).
• Ability to explain technical concepts to non-developers in English and French.

Responsibilities

• Build the platform's AI components: RAG, embeddings, ingestion pipelines, prompts and agent chains.
• Run and optimize LLM inference on NVIDIA GPUs: deploy and tune inference engines with quantization, batching, and KV cache.
• Operate models in production on customer hardware: lifecycle, versioning, monitoring, upgrades without downtime.
• Work on data architecture: vector stores, migrations, model selection and benchmarking.
• Own the product scope: roadmap, trade-offs, prioritization, production sign-off.

Skills

Python

FastAPI

Git

RAG pipeline

LLM inference

NVIDIA GPUs

PostgreSQL

MongoDB

Firebase

Docker

Linux

CUDA

Kubernetes

AWS

GCP

English

French

Tools

Docker

Kubernetes

AWS

GCP

Job description

The role

We're looking for an AI engineer who codes, ships and talks to customers. You own your scope, from roadmap to support, and you work directly with demanding enterprise customers in France and abroad.

Understand Tech is a startup of around ten people building an AI platform for enterprises: assistants, agents, business apps, and NVIDIA servers delivered to the customer's site (AI-in-a-Box) for those who want to keep their data in-house. Our customers come to us for sovereignty, GDPR compliance and AI governance. We work with Dell and NVIDIA, and we serve international customers across industries, from semiconductors to professional services.

We're now expanding to help our customers build their own AI factories: on-premises infrastructure, platform and applications, all running in their walls. That's where this role comes in: the models you serve, and how fast and reliably they run on the customer's hardware, are a big part of what we deliver. We're growing fast, which makes for a stimulating environment: things move all the time, we adapt constantly, and there's real room to grow.

What you'll do

• Build the platform's AI components: RAG, embeddings, document ingestion pipelines, prompts and agent chains, answer evaluation.
• Run and optimize LLM inference on NVIDIA GPUs: deploy and tune inference engines (vLLM, TensorRT-LLM, NIM, Ollama), quantization, batching, KV cache, throughput and latency on single machines and multi-GPU racks.
• Operate models in production on customer hardware: model lifecycle, versioning, monitoring, GPU usage, upgrades without downtime.
• Work on the data architecture: vector stores, migration to pgvector, model selection (Nemotron, Qwen, gpt-oss, etc.) and benchmarking.
• Own your product: roadmap, trade-offs, GitHub ticket prioritization, sign-off on what goes to production. Current example: our TCG (Test Case Generator), which generates test cases from technical specs.
• Handle technical support for a few accounts: reproduce a bug reported on Slack, find the root cause, fix it or get it fixed, reply to the customer.
• Build demos and POCs for prospects with the sales team.
• Write the documentation for what you ship, for the team and for customers.

Your profile

• You show ownership, autonomy and a sense of urgency.
• Solid Python, FastAPI or equivalent, Git every day.
• You've built a RAG pipeline end to end: chunking, embeddings, vector database, evaluation.
• You've served LLMs yourself on NVIDIA GPUs with vLLM or an equivalent engine, and you know how to make them faster: quantization, batching, context length, memory.
• You know PostgreSQL and at least one NoSQL database (MongoDB, Firebase).
• Docker without pain. Linux, CUDA drivers and GPU troubleshooting don't scare you. Kubernetes, AWS or GCP are a plus.
• You can explain a technical problem to a non-developer customer, in French and in English (a good share of our customers write in English).
• You go all in. On your scope, you decide what we build and you're accountable for it; you don't just code tickets. And you like being challenged on your decisions.

Nice to have: software testing or test generation, fine-tuning (LoRA, etc.), experience with Blackwell or Grace Blackwell hardware.

Details

Contract: Permanent (CDI)

Location: Frontignan

Work mode: On-site

Salary: To be discussed

Start date: As soon as possible

Process: A conversation with key leaders, a hands-on case study, a meeting with the tech team