ML Engineer (US Market Research SaaS) to $5000

Хелоу, люба спільното \0/

Ми продовжуємо нарощувати команду американського Market Research SaaS, що аналізує поведінку юзерів під час digital опитувань. По суті, це платформа, де бізнес може засетапити своє опитування та отримати швидкий результат, очищений від ботів + отримати advanced behavioural analytics від нашої моделі.

Вам сподобається, якщо ви:
- самодостатні та организовані.
- хочете автономію та відповідальність за свою ділянку.
- вам дійсно цікаво те, чим ви займаєтесь.

На цьому я дозволю перемкнутись на англійську, бо далі конкретніше.

About the role

Build and own ML models that predict survey respondent certainty and detect fraud from behavioral signals — mouse movement, click patterns, timing, browser/OS data. You'll be the sole ML engineer maintaining a production hybrid CNN+tabular certainty model (already in client deliverables) and building the new bot detection model. You'll work directly with a US-based AI scientist who provides strategic direction.
 

What you'll work on

• Improve the hybrid 1D CNN (mouse trajectories) + MLP (tabular features) certainty model — error analysis, hyperparameter tuning, augmentation, loss experiments 
• Build the bot detection classifier — binary human/bot model reusing behavioral pipeline infrastructure 
• Run segment-level evaluations — per-survey quality checks, precision/recall optimization, calibration analysis 
• Track experiments in MLflow — every run logged with parameters, metrics, dataset versions 
• Process raw time-series — mouse trajectories as [T, 7] tensors (x, y, speed, acceleration, dt, dx, dy) 
• Document to audit standard — code, dataframes, notebooks, reproducible results with artifact links
 

What we're looking for

• Strong PyTorch — built and trained CNNs from scratch, not just fine-tuned 
• Tabular ML experience — LightGBM, XGBoost, or gradient boosted trees 
• Rigorous evaluation mindset — calibration, segment analysis, label quality, beyond single-metric optimization 
• Time-series/sequence data experience — temporal patterns in any domain (NLP, audio, sensors, clickstream) 
• Imbalanced classification expertise — precision/recall tradeoffs, threshold tuning, AUC-ROC 
• MLflow or similar (W&B, Neptune) 
• Self-directed — you're the only ML person on execution, strategic guidance from US AI scientist 
• Auditable work — advisor reviews regularly, notebooks/scripts must be clean

• Clean verbal English
 

Nice to have

Behavioral/clickstream analysis • time-series augmentation • fraud/anomaly detection • model serving • survey methodology/psychometrics
 

Stack

Python • PyTorch • LightGBM • MLflow • HDF5 • PostgreSQL
 

What this is NOT about

Not LLM/GenAI — classical ML + deep learning on structured behavioral data 
Not research-only — models ship in client deliverables, production quality required 
Not team lead — hands-on role with strategic support from US AI scientist

The Team

• Product Manager
• 2x Fullstack Devs
• Anti-Fraud Automation QA
• DevOps
   

We offer

• 20 paid day-offs per year.
• 10 US/UA holidays.
• Flexible schedule and fully remote work from anywhere with reliable overlap to CET.
• Senior team members and strong AI Scientist Advisor.
• Minimal bureaucracy, high authonomy and trust, fast decisions.