← Все вакансии/Higgsfield AI
OfficeAlmaty

Product ML Engineer

HA
Higgsfield AI
Формат
Office
О роли

Описание вакансии

About the company
  • Higgsfield AI — fastest-scaling generative AI company, $500M annual revenue run rate, 25M+ users, 6M+ generations per day.
Responsibilities
  • Build ranking systems for content surfaces: effects and presets, templates, models, prompts, and what to show a user next after a generation completes.
  • Solve cold start for new users with one onboarding signal and no history.
  • Handle position bias, popularity feedback loops, exploration vs exploitation.
  • Rank against the objective of a completed, kept, shared generation, not the click.
  • Own incrementality on offers: discounts, credit vouchers, trials, plan upgrade prompts, win-back campaigns.
  • Model uplift, not propensity.
  • Design randomized holdouts for uplift training and measurement.
  • Respect guardrails: margin and abuse signals as constraints on the objective.
  • Work with Legal on personalization rules.
  • Build churn and downgrade prediction for subscribers, repeat-purchase propensity for credit-pack buyers.
  • Watch for leakage in churn models.
  • Pair every propensity model with an intervention and an experiment.
  • Turn user generations into features: use-case and intent labels over prompts, input assets, output assets, model and parameters.
  • Mine failed, abandoned and refunded generations.
  • Own models end to end: problem framing, features, training, offline evaluation, serving, monitoring, retraining.
  • Ship behind experiments; monitor for drift and degradation.
Requirements
  • Experience shipping ML that served live user traffic and changed a business metric.
  • Depth in at least two of: recommender systems / learning-to-rank, uplift & causal ML, churn or propensity modelling, real-time personalization.
  • Causal literacy: randomized holdouts, incrementality, Qini/uplift evaluation, selection effects.
  • Strong Python and SQL; gradient boosting and neural ranking; feature pipelines, training/serving skew, latency budgets, retraining cadence.
  • Product judgment: name the decision and the metric before picking the model.
  • Comfort with images and video as data.
  • Pragmatism: heuristic shipped this month beats a two-quarter platform.
  • Английский B2+.
Conditions
  • On-site role in Almaty, Kazakhstan.
Стек и навыки

С чем работаем