Remote job
Ingénieur(e) IA en déploiement avancé, Travel Intelligence// Forward Deployed AI Engineer, Travel Intelligence
Job details
About this role
Role overview Deploy and adapt machine learning models for enterprise travel use cases, working directly within partner environments from problem definition through production monitoring. You will lead model training and evaluation on site or in approved infrastructure, connect partner teams with research colleagues, and use evidence from real-world performance to guide improvements.
Responsibilities - Work with partner data and business teams to define priority tasks, build fixed evaluation sets, and agree on measures such as correctness, policy compliance, cost, and response time. - Create automated evaluators with targeted human review, establish baselines, and design experiments that clarify the causes of model performance. - Prepare partner data for training through extraction, redaction, deduplication, labeling, and train/test splitting, using only data needed for agreed use cases. - Fine-tune models with the research team and run training on partner or approved cloud and GPU infrastructure under the partner’s access and security rules. - Move models into production when evaluation supports it, then monitor quality, latency, and cost and use production results to improve them. - Present methods, results, and tradeoffs to senior partner stakeholders; share field findings with research and turn deployment lessons into reusable tools. - Maintain strong privacy and security practices and provide documentation for partner risk, security, and compliance reviews.
Requirements - Experience completing the full post-training lifecycle for open-weight models, from data preparation and fine-tuning through evaluation and serving. - Practical supervised fine-tuning experience and experience with preference optimization or reinforcement learning, including parameter-efficient methods such as LoRA. - Strong Python skills and familiarity with tools such as PyTorch, Hugging Face, and vLLM; able to run workloads on rented or customer-owned GPUs. - Experience building evaluation sets and evaluators from messy real-world data and comparing models on quality, cost, and latency. - Habit of using held-out data, baselines, and ablations, and of explaining both findings and limitations clearly. - At least two embedded deployments with enterprise customers, including work within their systems, data controls, and security reviews.
Nice to have - The source emphasizes confidence presenting to senior data and technology stakeholders and the ability to challenge partner assumptions constructively.