URAP

AI Visual Narratives

J. Miguel Villas-Boas, Professor  
Business, Haas School  

Open. Apprentices needed for the fall semester. Enter your application online beginning August 21st. The deadline to apply is Monday, August 31st, 4pm.

We are looking for 1-2 motivated undergraduate students to join an early-stage research project in the Quantitative Marketing group at Berkeley Haas.
The project broadly studies how generative AI systems (eg. LLMs, VLMs, and image-generation models) understand and produce coherent visual narratives. Potential directions include multimodal reasoning, controllable generation, AI evaluation, and visual storytelling.
Because the project is at an early stage, specific research questions and tasks may evolve as we learn from initial experiments. This role is best suited for students who enjoy exploring open-ended problems, learning new technical tools independently, and helping shape a research project from the beginning.

Application Process
Applicants should submit their application through the URAP portal and have weekly availability. If available, include a GitHub profile or example of previous technical work in the application.
Selected applicants will complete a small research exercise, submit a resume, and attend a short interview. The exercise will emphasize reasoning, communication, and willingness to learn.
Given the competitiveness of past application cycles, applicants who are also interested in a lower-hours, flexible research role are encouraged to also consider applying to “AI and Market Outcomes (Flexible Research Position).”

Role: Possible Responsibilities
- Help develop and test AI and multimodal-model workflows.
- Construct, clean, and evaluate research datasets.
- Audit machine-generated outputs for quality, consistency, and failure cases.
- Run controlled experiments and summarize results using Python.
- Read and synthesize relevant computer science and machine-learning literature.
- Maintain clear documentation and reproducible code.
- Explore new methods, models, and evaluation approaches as the project develops.
- Keep clean documentations of your work on GitHub.
If two students are selected, responsibilities may be divided between data/benchmark development and model experimentation. Students are expected to pay for their own computing and storage services. Due to the long publication cycle of business / economic projects, most of our URAP programs don't grant authorship.

Working with AI Tools
- We encourage students to use tools such as ChatGPT and coding assistants as part of the research process. At the same time, responsible AI collaboration is central to this position.
- Students should be able to understand, verify, and explain the code, analysis, and writing they submit, even when AI tools helped produce them. We value thoughtful questioning and careful validation over simply accepting or copying an AI-generated answer.

What You’ll Gain
- Hands-on experience with modern generative and multimodal AI systems.
- Exposure to the process of developing an early-stage research project.
- Experience with research coding, model evaluation, and reproducible experimentation.
- Close collaboration with graduate students in economics and marketing
- For students who demonstrate sustained engagement: mentorship on building an academic CV and graduate school application Q&A (experienced mentors from computer science / economics background available)
- Potential to contribute to multiple projects if bandwidth permits

Qualifications: Qualifications (required)
- Responsibility, attention to detail, and clear communication.
- Working knowledge of Python or a strong willingness to improve quickly.
- Experience in machine learning, generative AI, computer vision, NLP, or data science.
- Comfort learning unfamiliar tools and concepts independently.
- Ability to critically evaluate AI-generated code and explanations.
- Willingness to ask questions and communicate when uncertain.

Preferred
- Relevant coursework or projects in programming, data science, machine learning, NLP, or computer vision.
- Previous research or substantial technical project experience.
- Prior experience with fine-tuning.
- Experience with Git, PyTorch, Hugging Face, model APIs, or data-processing pipelines.

Day-to-day supervisor for this project: Debbie Liang, Graduate Student

Hours: 9-11 hrs

Off-Campus Research Site: or virtual.

Return to Project List