Research Scientist / AI Engineer Intern — Temporal Intelligence

Abel AI, Inc.; United States; Presencial; USD 18 000 - 30 000; Tempo integral

Resumo da oferta

Abel AI is seeking interns to work on focused research or engineering projects at the core of its temporal intelligence stack. You will collaborate with researchers and engineers to scale foundation models, develop multimodal temporal models, and build agentic forecasting systems.

Ideal candidates are pursuing a Bachelors, Masters, or PhD in CS, ML, or related fields with strong ML/DL fundamentals and hands-on PyTorch and Python experience. Publications and large-scale ML experience are a plus.

Qualificações

• Pursuing a Bachelor’s, Master’s, or PhD in CS, Applied Math, Statistics, EE, ML or related fields.
• Strong ML and DL fundamentals required.
• Hands-on PyTorch experience and solid Python programming skills.
• Experience with time-series forecasting, LLMs, multimodal learning, and large-scale ML systems.

Responsabilidades

• Own a focused research or engineering project and collaborate with our research and engineering teams.
• Contribute to scaling temporal foundation models and evaluate on large-scale datasets.
• Develop multimodal temporal models that combine text, time series, and structured signals.
• Build agentic forecasting systems that decompose questions and reason under uncertainty.
• Explore training and adaptation methods for temporal foundation models.
• Investigate causal discovery, causal inference and counterfactual reasoning for forecasting.
• Improve inference pipelines for low-latency forecasting across many tasks.
• Design experiments and benchmarks to understand model strengths and limitations.

Conhecimentos

ML fundamentals

Python programming

Time-series forecasting

Multimodal learning

Causal inference

LLMs / foundation models

Agentic AI systems

Large-scale ML systems

Formação académica

Bachelor’s/Master’s/PhD in CS, Math, Stats, EE, ML

Ferramentas

PyTorch

Python

Descrição da oferta de emprego

Own a focused research or engineering project at the core of Abel’s temporal intelligence stack.

Our mission

The world is always changing. The key is to understand the dynamics behind observations, predict how they will evolve, and decide what to do next.

We aim to build temporal superintelligence: AI systems that learn from massive-scale observations, understand temporal dynamics, forecast future outcomes, and reason about decisions under uncertainty.

Our goal is to push AI beyond static pattern recognition toward systems that can answer: What is happening? Why is it happening? What will happen next? And what should we do about it?

What you’ll work on

As an intern, you will own a focused research or engineering project and work closely with our research and engineering team. Depending on your background, you may work on:

• Scaling temporal foundation models. Help train and evaluate foundation models on datasets scaling toward 10T+ observations, and study how model, data, and compute scaling affect forecasting performance.
• Multimodal temporal modeling. Develop models that combine text, time series, tabular data, events, and other structured signals for forecasting and interpretation.
• Agentic forecasting. Build agents that decompose forecasting questions, retrieve relevant information, invoke forecasting models and tools, reason about uncertainty, and synthesize answers.
• Foundation model training and adaptation. Explore pre-training, post-training, fine-tuning, and adaptation methods for temporal and multimodal foundation models.
• Causal and counterfactual reasoning. Investigate how causal discovery, causal inference, and counterfactual reasoning can improve forecasting and decision-making.
• Efficient inference. Improve model serving and inference pipelines for scalable, low-latency forecasting across large datasets and many forecasting tasks.
• Research and evaluation. Design experiments and benchmarks to understand where temporal models succeed, where they fail, and how to make them more general and reliable.

What we look for

• Currently pursuing a Bachelor’s, Master’s, or PhD in Computer Science, Applied Mathematics, Statistics, Electrical Engineering, Machine Learning, or a related field.
• Strong understanding of machine learning and deep learning fundamentals.
• Hands-on experience with PyTorch and strong Python programming skills.
• Experience with at least one of the following:
  • Time-series forecasting or representation learning
  • LLMs and foundation models
  • Pre-training or post-training
  • Multimodal learning
  • Causal discovery or causal inference
  • Agentic AI systems
  • Large-scale ML systems
• Ability to independently implement research ideas, design experiments, analyze results, and iterate quickly.
• Curiosity about time, dynamics, forecasting, causality, and decision-making.
• Experience training or fine-tuning LLMs, time-series foundation models, or other large-scale models.
• Experience working with large-scale time-series or tabular datasets.
• Experience with distributed training, GPU clusters, efficient inference, or ML systems optimization.
• Hands-on experience building AI agents involving tool use, retrieval, planning, code execution, or model orchestration.
• Research experience in forecasting, temporal modeling, causal learning, multimodal learning, or generative modeling.
• Publications at venues such as NeurIPS, ICML, ICLR, KDD, AAAI, AISTATS, or related conferences, or meaningful open-source contributions.

Publications are a plus, not a requirement. We care more about your ability to understand difficult problems, build things, run rigorous experiments, and learn quickly.

What you’ll get

You will work on problems at the intersection of foundation models, time series, multimodal AI, causal learning, and agentic systems.

This is not an internship where you spend the summer on an isolated toy problem. You will work on a focused problem connected to our core models and infrastructure, with the opportunity to contribute to research, open-source systems, and production models.

Today’s foundation models are remarkably good at understanding what has already been written. We want to build models that understand what happens next.