20-Min Screening Interview

Q: Tell me about your internship at Creek & River.

I worked as a Market Research Intern in Tokyo, where I analyzed multilingual customer inquiry data. I used Python, R, and AI-assisted translation tools to clean and organize unstructured customer inquiries into standardized categories. I then analyzed trends, created visualizations in R, and presented insights to a 30-person team to help improve how clients were matched with companies. More recently, I've also been helping develop AI-assisted workflows to automate survey analysis.

Q: What was the AI automation project?

During my internship, our team developed an AI-assisted chatbot using Claude to automate survey and sales call analysis. Instead of manually reviewing large datasets, users could upload a dataset and ask questions in natural language, such as "Analyze the sales call data for returning customers only." The chatbot would then generate summaries and insights based on that request. My role was to help iterate on the workflow, evaluate how well the outputs addressed the prompt, and improve the overall user experience before handing the tool off.

Q: Why did you build it?

We wanted to reduce the time spent manually analyzing survey and sales call data. Many analyses followed a similar pattern, so we explored whether users could simply ask questions in natural language instead of manually filtering data and writing reports.

Q: What AI tools did you use?

We primarily used AI-assisted translation tools to work with multilingual datasets, and more recently we've explored AI tools to help categorize survey responses and assist with report generation. The goal wasn't to replace human analysis, but to reduce repetitive manual work.

Can you tell me about the chatbot you built?

During my internship, my intern team and I built an AI-powered chatbot using Claude to automate survey and sales call analysis. The goal was to reduce the amount of manual work required to analyze large datasets.

Users could upload a dataset and ask questions in natural language, such as "Analyze the sales call data for returning customers only," and the chatbot would generate summaries and insights based on that request.

We iterated on the workflow about three times, testing different prompts, evaluating the quality of the responses, and refining the user experience before handing the project off.

Why Claude?

Claude was well-suited because it could interpret natural language prompts and summarize large amounts of information. The goal wasn't to build our own AI model, but to leverage an existing LLM to make data analysis more accessible and efficient.

What kinds of prompts did you test?

Examples:

• Analyze sales call data for returning customers only.

• Summarize the most common customer complaints.

• Identify recurring themes across survey responses.

• Compare responses by customer segment.

• Generate a summary report.

Q: What exactly did you contribute?

I helped process survey data, organize responses, explore methods for categorizing recurring themes, and support the development of an AI-assisted workflow for generating reports more efficiently.

Q: Why are you interested in AI automation?

My interest in AI originally came from working on a machine translation project for low-resource languages. Through that project and my internship, I realized that AI isn't only useful for building models—it can also automate repetitive tasks and help people work more efficiently. That's what attracted me to this internship.

Q: Tell me about Lost in Translation.

I led a project that fine-tuned transformer-based neural machine translation models for low-resource languages. We compared different attention mechanisms using BLEU and Perplexity metrics and found that Cross Attention produced the best translation performance.

Q: Why SprintRay?

What interests me most is the opportunity to apply AI to improve real business processes. I like that this internship focuses on understanding existing finance workflows first and then identifying opportunities where AI and automation can reduce repetitive work and improve decision-making. That practical approach really aligns with what I enjoyed during my internship at Creek & River.

Q: What is your experience with AI?

I've used AI in several different ways: machine learning through my NLP project, AI-assisted translation for multilingual datasets during my internship, and more recently helping develop AI-assisted workflows to automate survey analysis. Those experiences have shown me how AI can support both technical modeling and day-to-day business processes.

Q: Do you know SQL?

Yes. I've used SQL through coursework and have continued practicing it independently. Most of my internship work was done in Python and R, but I'm comfortable with SQL fundamentals such as joins, filtering, grouping, and aggregations, and I'm continuing to strengthen those skills.

What challenges did you encounter?

Situation

• During my internship, our team spent a significant amount of time manually analyzing survey and sales call data to identify trends and prepare reports.

Task

• We wanted to find a way to make that process more efficient while still producing useful insights.

Action

• My intern team and I proposed building an AI-assisted chatbot using Claude that would allow users to upload datasets and ask questions in natural language.

• We developed the workflow, tested different prompts, and iterated on the design about three times to improve the quality of the outputs and make the tool easier to use.

Result

• We delivered a working prototype that automated portions of the survey analysis process and reduced the amount of manual effort required to generate insights. The experience taught me how important it is to understand users' workflows and iterate based on testing rather than assuming the first solution is the best.

"If I gave you a finance dataset tomorrow, how would you use AI to automate part of the analysis?"

"I'd first understand what the finance team currently does manually. Then I'd identify repetitive tasks, such as summarizing transaction data, categorizing expenses, or generating recurring reports. Rather than replacing human review, I'd use an LLM like Claude to automate the initial analysis or draft reports, allowing employees to focus on validating results and making decisions."

Tell me about yourself

Tell me about yourself

• Current student

  • UCLA Cognitive Science major

  • Minor in Statistics & Data Science

• Interest

  • Passionate about using AI and data to solve real-world business problems

  • Interested in AI automation and improving workflows

• Internship

  • Market Research Intern at Creek & River in Tokyo

  • Worked with multilingual customer inquiry datasets using Python, R, and AI-assisted tools

  • Analyzed data and presented insights to a 30-person team

• AI Automation

  • Helped develop an AI-powered chatbot using Claude

  • Users could upload datasets and ask questions in natural language

  • Iterated on the workflow three times to improve usability before handing it off

• Machine Learning

  • Led an NLP project on low-resource language translation

  • Compared transformer attention mechanisms using BLEU and Perplexity metrics

• Why SprintRay

  • Excited about applying AI to automate business processes

  • Interested in learning how AI can improve finance workflows and support better decision-making

What is a weakness of yours?

One challenge I faced early in my internship was adjusting to a professional work environment. At first, I was hesitant to ask questions because I wanted to solve everything on my own. I realized that spending too much time stuck wasn't productive, so I started asking clarifying questions earlier. That helped me complete tasks more efficiently and taught me that asking thoughtful questions is an important part of collaborating in the workplace.

What did you learn from building it?

One thing I learned is that building an AI tool isn't just about connecting it to an LLM. A lot of the work is understanding the user's workflow, designing clear prompts, testing different scenarios, and refining the experience so the outputs are actually useful. Through our iterations, I realized that small changes to the workflow could have a big impact on how effectively users could analyze their data.