Generative AI Research Intern

Appier is a software-as-a-service (SaaS) company that uses artificial intelligence (AI) to power business decision-making. Founded in 2012 with a vision of democratizing AI, Appier's mission is to turn AI into ROI by making software intelligent. Appier now has 17 offices across APAC, Europe, and the U.S., and is listed on the Tokyo Stock Exchange (Ticker number: 4180). Visit www.appier.com for more information.

As a Research Intern, you will work alongside our research scientists at the frontier of generative and agentic AI: Large Language Models (LLMs), Vision-Language Models (VLMs), and AI agents that reason, plan, and use tools to solve real-world problems.

You will own one focused research project end-to-end: from literature survey and hypothesis, to implementation, evaluation, and a working prototype. Strong projects often lead to a paper submission, an internal benchmark or tool that the team keeps using, or a feature that ships into Appier products. You will have a dedicated mentor, weekly 1:1s, access to GPU resources and frontier models, and a seat in our research reading group.

This is a paid internship. We look for people who want to keep building with us: top performers are considered for full-time Research Scientist / Research Engineer roles.

You'll go deep on one of the following (final topic is scoped with your mentor based on your interests):

• Agentic AI: reasoning, planning, tool use, memory, or multi-agent collaboration on top of LLMs/VLMs.
• Post-training: SFT, RLHF / RL with verifiable rewards, or preference optimization to improve capability and reliability.
• Efficiency & test-time scaling: making models faster, cheaper, or smarter with more inference compute.
• Evaluation: designing benchmarks and evals that actually predict real-world agent behavior.
• Multimodal intelligence: VLMs applied to Appier's marketing, creative, and commerce data.

• →Survey relevant literature and turn it into a concrete, testable research plan with your mentor.
• →Implement, train, and evaluate models or agent pipelines in Python/PyTorch.
• →Build a working prototype or demo that lets the team judge whether the idea holds up in practice.
• →Run rigorous experiments: clear baselines, ablations, honest error analysis.
• →Share progress in weekly syncs and present your final results to the research team.

• Currently enrolled in a Bachelor's (junior/senior year) or Master's program in Computer Science, Electrical Engineering, Mathematics, Statistics, or a related field.
• Solid foundation in machine learning and deep learning, and familiarity with how modern foundation models work.
• Proficient in Python and comfortable with PyTorch; you can read a paper's repo, run it, and modify it.
• Hands-on experience building something with LLMs: fine-tuning, RAG, loop engineering, agent frameworks (tool use/function calling), or a side project you can walk us through.
• Curiosity, self-direction, and the ability to work comfortably through ambiguity and drive a project with weekly guidance.
• Clear communication and a collaborative attitude.
• Available for at least 3 months, minimum 3 days per week (full-time over summer/winter break preferred).

• Research experience in a lab, a submission to a top AI/ML venue (NeurIPS, ICML, ICLR, CVPR, ACL, EMNLP), or a course project of comparable depth.
• Experience with the modern LLM stack: Hugging Face, vLLM/SGLang, LoRA/PEFT, TRL/verl, LangGraph or similar agent frameworks.
• Open-source contributions, competition results (Kaggle, ML challenges), or a public repo/blog showing your work.
• Fluency with AI-assisted coding workflows.
• Experience with multi-GPU or distributed training.

• A dedicated research mentor and clear project scope from week one.
• GPU compute and access to frontier models and internal data.
• Real product exposure: your prototype gets used and critiqued by the people who ship.
• Support toward publication or open-source release when the work merits it.
• A fast track to full-time consideration.

Please include your CV/transcript plus links to any of the following: GitHub, publications, personal site, or a short description of a project you're proud of. A concrete project beats a long list of course names.

#LI-JC1