Forward Deployed AI Engineer, Travel Intelligence
Hopper, Canada - Remote +3
Hopper has this job in 1 other place
The role
Hopper powers travel for some of the world’s largest banks, airlines and travel companies, and our Research team builds travel-specialized models. As a Forward Deployed AI Engineer you take that work into a partner’s environment and make it theirs: building the evaluation with their team, preparing their data, training and evaluating the model, and putting it into production where it delivers measurable results. You work in a small forward-deployed team as the person who leads model training on site, across servicing, loyalty, commerce and more, and you are the bridge between our partners and Research.
What would your day-to-day look like
- Work with a partner’s data and business teams to turn their real tasks into a frozen, held-out benchmark across their priority use cases
- Define what success means for each task, covering task completion, policy adherence, cost and response time, and build the automated graders with human spot checks
- Establish baselines so results can be compared fairly
- Prepare training data inside the partner’s environment through extraction, redaction, deduplication, labelling and train/test splits, using only the data the agreed tasks need
- Run post-training in partnership with Research, applying modern fine-tuning methods to partner use cases
- Run training on the partner’s cloud and GPUs, or in an environment they approve, under their security and access rules
- Design experiments that show partners clearly what drives the results
- Take models from evaluation into production based on evidence, then monitor quality, latency and cost once they are live
- Use production results to keep improving the model
- Present results, method and trade-offs to partner data and technology leaders, and defend them
- Work closely with Research, feeding back what you learn in the field
- Turn each deployment into reusable tooling so the next partner goes faster
- Uphold strict data handling, privacy and security standards in every partner environment, and support partner model-risk, security and compliance reviews with clear documentation of data, training and evaluation
An ideal candidate has
- Run the full post-training loop end to end themselves, from data preparation through fine-tuning an open-weight model, evaluation and serving, rather than only calling a vendor’s fine-tuning service
- Working experience with supervised fine-tuning and at least one of preference optimization or reinforcement learning, including parameter-efficient methods such as LoRA
- Strong Python and command of the standard training stack, for example PyTorch, Hugging Face and vLLM, with comfort running jobs on rented or customer GPUs