Mechanistic Interpretability Jobs in Georgia (NOW HIRING)

Equifax is where you can power your possible. If you want to achieve your true potential, chart new paths, develop new skills, collaborate with bright minds, and make a meaningful impact, we want to hear from you.
We are seeking an AI Research Engineer to develop Transformer-based models for structured and time-series credit risk data, with a strong focus on representation learning and interpretable AI attributes.
Your mission is to help build the foundation for a long-horizon initiative that begins with interpretable feature learning and expands into discriminative and generative modeling. This role is ideal for a scientist who is strong in deep learning fundamentals, curious about how models learn internal representations, and excited to work on the next generation of transformer applications for structured data under senior technical mentorship.
Equifax has a hybrid work schedule that allows for two days of remote work (Monday and Friday) with 3 days onsite (Tuesday thru Thursday) every week.
This role reports to our office Alpharetta, GA or our Midtown (OAC, Atlanta) office.
This position does not offer immigration sponsorship (current or future) including F-1 STEM OPT extension support.
This is a direct-hire role and is not open to C2C or vendors.
What you will do

• Advance Experimental Research: Build and experiment with transformer-based models specifically for structured and credit time-series data, pushing the boundaries of model performance and capability.
• Analyze Internal Representations: Investigate and interpret learned representations (embeddings, latent spaces, attention patterns) to uncover how the model encodes complex financial concepts.
• Execute Rigorous Experimentation: Conduct ablations, hyperparameter sweeps, and controlled experiments to validate hypotheses on model behavior and training dynamics.
• Develop Research Prototypes: Train, evaluate, and debug deep learning models using PyTorch/TensorFlow, creating high-fidelity prototypes that provide the conceptual and architectural blueprint for the engineering team.
• Collaborate on Integration: Partner with our internal ML Engineering team to ensure your research prototypes are successfully integrated into production pipelines.
• Drive Strategic Expansion: Work with senior technical leadership to extend our core models beyond interpretability into broader discriminative and generative modeling architectures.
• Stay at the Vanguard: Maintain deep currency with modern deep learning techniques, including sequence-to-sequence, diffusion models, and generative approaches.

What experience you need

• Education & Experience: A PhD in ML/AI/CS/EE or a related quantitative field with 3+ years of relevant experience; OR an MS with 5+ years of relevant industry/research experience.
• Deep Learning Foundations: Strong, demonstrated foundation in Transformer architectures, attention mechanisms, and sequence modeling.
• Mathematical Maturity: A deep, working knowledge of linear algebra, statistics, and probability-the foundational mathematics required to characterize model behavior, evaluate representation similarity (e.g., Kernel CCA), and derive insights from internal model activations.
• Representation Learning: Experience analyzing and working with learned representations (latent spaces, embedding analysis, internal model states).
• Training Intuition: Strong technical intuition for deep learning training dynamics-specifically regarding stability, gradient behavior, and learning rate schedules.
• Programming Rigor: Ability to write clean, well-structured, and efficient Python code.
• Soft Skills: Demonstrated curiosity, technical ambition, and a desire to grow your research career under senior technical leadership.

What could set you apart

• Advanced Modeling: Experience with sequence-to-sequence models, diffusion models, or other generative modeling techniques.
• Deep Analysis: Experience analyzing or interpreting learned representations through techniques such as probing, attribution, or embedding visualization.
• Data Complexity: Experience with irregular time-series, missingness handling, or temporal embedding techniques.
• Mechanistic Interpretability: Familiarity with mechanistic interpretability concepts, such as sparse autoencoders, feature dictionaries, activation analysis, or attention-pattern interpretation.

#LI-AM2
#LI-Hybrid
We offer comprehensive compensation and healthcare packages, 401k matching, paid time off, and organizational growth potential through our online learning platform with guided career tracks.
Are you ready to power your possible? Apply today, and get started on a path toward an exciting new career at Equifax, where you can make a difference!
Primary Location:
USA-Atlanta JV White
USA-Atlanta-One-Atlantic-Center
Function:
Function - Data and Analytics
Schedule:
Full time