VP of Research (Machine Learning)
8Bit - Games Industry Recruitment
⚲ Warszawa
30 000 - 40 000 PLN brutto (UoP)
Wymagania
- Machine Learning
- Python
Opis stanowiska
Our client, an early-stage AI company, is hiring a VP of Research (Machine Learning) to lead the intelligence and research direction behind a proactive AI assistant.
Billions of people still run their day-to-day through tools that weren’t built with AI in mind – inboxes, notes, to-do lists. This product aims to change that, cutting the time people spend on routine tasks by roughly 90% through reliable, multi-step, tool-using AI.
You’d be the person deciding how the system reasons, learns, and gets evaluated, on a product already used at high frequency.
RESPONSIBILITIES
• Chart the research roadmap across memory, context handling, reasoning, planning and orchestration for the assistant’s core intelligence
• Call the shots on building custom model architecture vs. adapting existing open-source or commercial frontier models
• Build out evaluation systems that capture real-world reliability and safety, not just benchmark scores
• Treat alignment, safety and guardrails as core product decisions, not an afterthought
• Push technical exploration into areas like retrieval-augmented training, mixture-of-experts, distillation, multi-agent setups and multimodal input
• Work hand-in-hand with product and engineering to shape what the assistant can do early on
REQUIREMENTS
• Fluent in Python and PyTorch/JAX, comfortable running GPU training and inference at scale
• A track record of shipping or evolving ML systems that run in production, not just research prototypes
• Sharp instincts for how models fail, behave, and hold up over long time horizons
• A hands-on builder who cares more about what works in the real world than what’s theoretically elegant
• Able to make high-stakes calls with incomplete information, and live with them
• Genuinely obsessed with evaluation and correctness – how the system behaves today and months from now
• Operates like a founder: full ownership, not delegation
NICE TO HAVE
• Has taken a research function from zero to one inside a startup before
• Practical experience with retrieval-augmented training, MoE, distillation, multi-agent systems or multimodal models
• Has personally owned safety/guardrail strategy for a product already live with users
WHAT THEY OFFER
• Cash and equity compensation
• Remote-first setup with flexible hours as part of a distributed, global team
• Generous paid time off
• Company laptop provided
• A quick hiring process – 3, occasionally 4, interviews, with fast decisions afterwards
ABOUT THE COMPANY
They’re an early-stage AI company building a proactive assistant aimed at the 5+ billion people currently stuck using non-AI-native tools for everyday things – email, notes, tasks.
The focus is squarely on reliability: long-running workflows, persistent context, and tasks that actually get done, even though the underlying models aren’t fully deterministic.
Stage: Early-stage AI startup
Focus: Proactive AI assistant for everyday productivity
Work mode: Remote-first, distributed team
Billions of people still run their day-to-day through tools that weren’t built with AI in mind – inboxes, notes, to-do lists. This product aims to change that, cutting the time people spend on routine tasks by roughly 90% through reliable, multi-step, tool-using AI.
You’d be the person deciding how the system reasons, learns, and gets evaluated, on a product already used at high frequency.
RESPONSIBILITIES
• Chart the research roadmap across memory, context handling, reasoning, planning and orchestration for the assistant’s core intelligence
• Call the shots on building custom model architecture vs. adapting existing open-source or commercial frontier models
• Build out evaluation systems that capture real-world reliability and safety, not just benchmark scores
• Treat alignment, safety and guardrails as core product decisions, not an afterthought
• Push technical exploration into areas like retrieval-augmented training, mixture-of-experts, distillation, multi-agent setups and multimodal input
• Work hand-in-hand with product and engineering to shape what the assistant can do early on
REQUIREMENTS
• Fluent in Python and PyTorch/JAX, comfortable running GPU training and inference at scale
• A track record of shipping or evolving ML systems that run in production, not just research prototypes
• Sharp instincts for how models fail, behave, and hold up over long time horizons
• A hands-on builder who cares more about what works in the real world than what’s theoretically elegant
• Able to make high-stakes calls with incomplete information, and live with them
• Genuinely obsessed with evaluation and correctness – how the system behaves today and months from now
• Operates like a founder: full ownership, not delegation
NICE TO HAVE
• Has taken a research function from zero to one inside a startup before
• Practical experience with retrieval-augmented training, MoE, distillation, multi-agent systems or multimodal models
• Has personally owned safety/guardrail strategy for a product already live with users
WHAT THEY OFFER
• Cash and equity compensation
• Remote-first setup with flexible hours as part of a distributed, global team
• Generous paid time off
• Company laptop provided
• A quick hiring process – 3, occasionally 4, interviews, with fast decisions afterwards
ABOUT THE COMPANY
They’re an early-stage AI company building a proactive assistant aimed at the 5+ billion people currently stuck using non-AI-native tools for everyday things – email, notes, tasks.
The focus is squarely on reliability: long-running workflows, persistent context, and tasks that actually get done, even though the underlying models aren’t fully deterministic.
Stage: Early-stage AI startup
Focus: Proactive AI assistant for everyday productivity
Work mode: Remote-first, distributed team
🔍 Dekoder Ogłoszenia
🔴
early-stage AI company
Firma jest na wczesnym etapie rozwoju, co może oznaczać niestabilność, brak ugruntowanych procesów i potencjalnie mniejsze zasoby.
🔴
VP of Research (Machine Learning)
Stanowisko to może oznaczać zarówno dużą autonomię i wpływ, jak i ogromną odpowiedzialność za kierunek badań w nowym, niepewnym produkcie.
🔴
product already used at high frequency
Produkt jest już używany, ale 'wysoka częstotliwość' może oznaczać zarówno sukces, jak i problemy z wydajnością lub stabilnością, które wymagają natychmiastowych rozwiązań.
🔴
Call the shots on building custom model architecture vs. adapting existing open-source or commercial frontier models
Decyzje o architekturze modeli mogą być bardzo czasochłonne i kosztowne, a brak jasnych wytycznych może prowadzić do nieefektywnych wyborów.
🔴
A track record of shipping or evolving ML systems that run in production, not just research prototypes
Oczekuje się doświadczenia w praktycznym wdrażaniu i utrzymaniu systemów ML, co może oznaczać presję na szybkie dostarczanie działających rozwiązań, a nie tylko innowacje teoretyczne.