Job
- Level
- Erfahren
- Job Feld
- Data
- Anstellung
- Vollzeit
- Vertragsart
- Unbefristetes Dienstverhältnis
- Ort
- Wien
- Arbeitsmodell
- Hybrid, Onsite
Job Zusammenfassung
In dieser Rolle bist du verantwortlich für die Kuratierung von menschlichem Feedback und die Entwicklung hochwertiger synthetischer Daten, um die Leistung von Modellen durch innovative Evaluationsstudien und Verbesserungen maßgeblich zu steigern.
Job Technologien
Deine Rolle im Team
- In this foundational role, you will be the expert on what fuels our models: data.
- You will be responsible for the end-to-end lifecycle of our datasets, from curating nuanced human feedback on design quality to pioneering the use of machine learning for high-quality synthetic data generation.
- You will design the sophisticated human evaluation studies and experiments that measure our success, and you will be the key analyst who translates signals from automated evaluators into a deep understanding of our models' alignment with human taste and intent.
- Human Feedback Data Curation: Owning the processing, cleaning, and strategic curation of large-scale, subjective human feedback on design quality, which is the lifeblood of our models.
- Synthetic Data Generation: Using generative AI and machine learning techniques to create novel, high-quality synthetic data that augments our training sets and improves model capabilities.
- Alignment Analysis & Evaluation Design: Designing methods to analyze outputs from both human and automated systems to deeply understand and measure our models' alignment with user preferences.
- Design and build scalable pipelines for processing and curating large datasets of human design feedback.
- Develop ML models to generate high-quality synthetic data for training and fine-tuning.
- Own the design and implementation of human evaluation workflows, including creating guidelines and quality rubrics.
- Prepare datasets for automated evaluation systems and analyze their outputs to provide a robust signal on model performance and human alignment.
- Design and analyze experiments (including A/B tests) to measure the real-world impact of our models on design quality.
- Conduct deep-dive analyses into model performance to identify failure modes and guide future development.
Unsere Erwartungen an dich
Qualifikationen
- A strong aesthetic sense, with a background or demonstrated passion for visual design or human-computer interaction.
- Strong proficiency in Python and ML frameworks (e.g., PyTorch, TensorFlow).
- Solid understanding of statistical methods, including experimental design, A/B testing, and quality evaluation systems.
- Familiarity with data annotation tools is a plus.
Erfahrung
- Experience with generative AI and synthetic data generation is highly desirable.
Benefits
Gesundheit, Fitness & Fun
- 🤫Ruheräume
- ⚽️Tischkicker o. Ä.
- 🎮Gaming Room
- 🧠Psychische Gesundheitsv.
- 🚲Fahrradabstellplatz
- 🎳Team Events
- 🙂Gesundheitsförderung
Work-Life-Integration
Essen & Trinken
Mehr Netto
Job Standorte
Standort Wien
Ungargasse 37
1030 Wien
Österreich
Themen mit denen du dich im Job beschäftigst
Das ist dein Arbeitgeber
Canva Austria GmbH.
Wien
Empowering the world to design by Visual AI: By making complicated tech simple, Kaleido strives to enable individuals and businesses of all sizes to benefit from the recent advances in visual AI. Our tools simplify and accelerate workflows, foster creativity, and enable others to create new products. Since 2021, we are part of Canva. Our mission at Kaleido as part of Canva is to empower the world to design and since Canva launching in 2013, we have grown exponentially, amassing over 100 million monthly active users across 190 different countries and a team of over 3,000 people… and the best bit is that we’ve only achieved 1% of what we know we’re capable of.
Description
- Unternehmensgröße
- 50-249 Employees
- Gründungsjahr
- 2018
- Sprachen
- Englisch
- Unternehmenstyp
- Startup
- Arbeitsmodell
- Hybrid, Onsite
- Branche
- Internet, IT, Telekom
Research Engineer - Datasets
- Ort
- Wien
- Arbeitsmodell
- Hybrid, Onsite
