At Megagon Labs, our interns don’t just learn — they drive innovation. In this second volume of our intern experience series, three summer interns and one fall intern discuss their work on projects that advance multi-agent LLM systems, human-AI interaction, and ambiguity handling in NL2SQL.
From improving coordination and verification in agentic systems to designing benchmarks that help models reason through uncertainty, each intern advanced research toward more reliable and transparent AI — work that also supports the essential foundations of agentic AI for enterprises.
Read on to learn how they approached open research questions, collaborated with Megagon researchers, and gained insights that extend beyond the lab.
Want to read Volume 1? Find it here.
Yihao Hu
University: Duke University
Program: PhD in Computer Science
Project Description: Create a benchmark for users to evaluate how well a model can handle ambiguity in text-to-SQL tasks and help improve a model when it encounters ambiguous task instructions.
Mentors: Yanlin Feng and Naoki Otani
Were there any moments where your project took you in an unexpected direction? How did you handle it?
We did underestimate the time it takes to annotate, revise, and correct queries. It’s a normal pain for high-quality benchmarks, since annotation helps to ensure its quality, especially for studying ambiguity in text-to-SQL. We knew it would take time to complete it, but it took even longer. We accepted that reality, knowing that time was an investment in quality. It’s very important that we guarantee the quality of the benchmark.
What was a big challenge and a success you encountered during your internship?
With our task, it was difficult to define the taxonomy: categorizing and defining types of ambiguity in an organized way. It was something that we found had not been done before, at least not with a strong rationale behind the categorization. It took a lot of creativity and innovation to develop a comprehensive taxonomy for ambiguity. I have to give a lot of credit to Naoki Otani, an expert in NLP.
How did this experience shape your view of what you want in a research career?
At school, I worked on databases, primarily focusing on query processing and query optimization. This internship at Megagon shifted my focus toward natural language processing (NLP). NLP is something I have always wanted to explore so I could learn how it overlaps with databases, and with that combined understanding, create something unique.
I also gained new experiences, especially in NL2SQL, which is a current hot topic. This will help bring new ideas to my future research.
Would you recommend this internship to others? Why or why not?
Yes, and this was my second time interning with Megagon Labs. It has helped me grow. I was able to explore new research directions that were relevant to my previous work, broadening my knowledge and interdisciplinary experience. At Megagon, everyone is open to discussions and encourages collaboration. I’ve really enjoyed both my internships here.
What advice would you give to future interns, now that you have interned here twice?
Prepare beforehand, and get in touch with your mentor early. You only have three months and the projects are intensive. You need all the time and preparation you can get.
Tianyang Xu
University: Purdue University
Program: PhD in Computer Engineering
Project Description: Improving the accuracy and effectiveness of multi-agent LLM systems through structured context sharing and verification-aware planning.
Mentors: Dan Zhang
How was your experience constructing the research project?
I worked with Dan to formulate the project, which made it more exciting than working on an already established project. It was challenging to formulate the research project and problem in a way that innovated and stood out from all the work on LLMs and agents being produced. I believe ours ended up being the first of its kind.
Were there any moments where your project took you in an unexpected direction? How did you handle it?
Yes, it was important to be flexible with the project’s direction. My mentor gave me valuable advice on alternative paths to take. We discussed potential directions with each other to work toward another plan.
What was your biggest 'aha' moment during your internship?
We devised this new system, and then we considered its structure. The ‘aha’ moment came when we finally resolved many detailed issues and saw the system start to work with all the datasets. It was as if our efforts from the previous months had finally succeeded. It was a rewarding milestone, although we weren’t done with the project itself.
How did this experience shape your view of what you want to do in a research career?
I am more likely to work as a researcher in the future because I enjoyed working with others, generating new ideas, and handling projects and codebases. I loved that rewarding feeling of getting a system to work after numerous attempts.
Would you recommend this internship to others? Why or why not?
I do recommend this internship because it is very helpful. My mentor invested a significant amount of time and effort in helping me make this project work. Megagon’s staff are very nice, and the work environment is positive.
What advice would you give to future interns?
I recommend you start preparing for your project and consider the topics you will work on before beginning the internship. It will help things be more straightforward and help you formulate the project quickly. That way, you make the best use of your mentor’s collaboration time.
Steven He
University: Penn State University
Program: PhD in Informatics
Project Description: Improve transparency and controllability of an interactive multi-agent system by enabling more fine-grained planning and providing semantic feedback.
What was the biggest challenge in completing your project?
I have faced two challenges. At the beginning of the internship, I had to spend about two weeks learning the intricacies of multi-agent systems, because my PhD focus has been on NLP and human-centered AI. It was a challenge, but also a good opportunity to be up-to-date with a prevalent AI topic.
The second challenge was system deployment. It was more intricate than I expected, but fortunately, I had support. The entire internship has really enhanced my full-stack software development skills.
Is your project different from what you were researching during your PhD?
It is close to what I’m doing for my PhD, but it was still different enough to help me grow in computer science. What I’ve been pursuing at Penn State is human-centered and AI-focused, but applying that to multi-agent systems was new to me. Now, I would like to delve further into multi-agent systems and system improvement throughout the remainder of my PhD.
How has the support of your mentors or other staff influenced your experience?
Everyone is super supportive when I hit roadblocks in design and development. Working with Megagon mentors has also helped me communicate better, to problem-solve, and work together toward a common goal.
The overall experience at Megagon Labs was impressive and unforgettable because we are not only focusing on a product. I saw a good balance of academic rigor and industry development. The internship gave me the ability to think deeply about my research.
I’d definitely do the internship again.
Do you have any advice for future interns?
Be sure to balance work and your personal life. With your free time, explore the Bay Area. There is a lot to do: hikes, restaurants, city views, and a lot of culture.
At the office, don’t be afraid to speak up. People will support you if you have questions.
Yue Zhang
University: University of Texas at Dallas
Program: PhD in Computer Science
Project Description: Exploration on how factors like table size, query complexity, and schema quality impact LLM performance on text-rich tabular data, which revealed key trade-offs between accuracy, flexibility, and robustness across SQL, LLM, and hybrid methods.
Mentors: Seiji Maekawa and Nikita Bhutani
What surprised you most about doing research in an industry lab compared to academia?
At Megagon Labs, we are applying the research to production, versus just doing the research and having it remain a high-level idea. In academia, it’s more theoretical, but here there’s more effort and thought put into the actual application of the research.
A second aspect that stood out is that there are many more resources available to me at Megagon Labs than at school.
What skill (technical or otherwise) did you improve the most during your internship?
During my internship, I identified some issues with my previous approach to research. In the past, I often prioritized novelty when selecting research topics, but did not systematically assess feasibility by conducting small-scale preliminary experiments. As a result, I sometimes encountered challenges later in the process or discovered fundamental issues with the direction I chose.
Through the internship, I learned the importance of validating ideas early on through pilot studies and feasibility analyses, which helps ensure that the research is both innovative and grounded.
I also learned to communicate and listen better. I am more mature in handling disagreements over research topics. I was able to witness the coordination and effort researchers invested in order to streamline understanding. I’ve changed and improved the style and tone of my communication.
If you had to describe your internship in three words, what would they be and why?
Happy – Everyone was so fun here, and the atmosphere was nice. I’ve felt very happy being in the office during my internship.
Research – During this internship, I conducted research and focused on a specific paper. Other internships don’t have the same focus. If you want to have a research-focused internship, come to Megagon.
Communication – My communication improved significantly, an essential skill for my career aspirations. I learned to communicate in a gentle manner and have healthy conversations, even when disagreements arise. I gained personal growth in a professional space.
Would you recommend this internship to others? Why or why not?
I would definitely recommend this internship to my friends because it is an entirely research-based position in an industry setting. Most PhD students go to big companies, but they often don’t get to conduct research or are only involved in a small portion of projects, without being able to formulate and execute a project from beginning to end. I also appreciated that my mentors were very hands-on and willing to support me with any blockers or questions immediately.
Do you have any advice for future interns?
Connect with your mentor before you start the internship. It would have saved a lot of time and given me a stronger start.
During your internship, connect more with your mentor so they can help redirect you if you are on the wrong track or so you can have someone to help think through methods or roadblocks.
Written by Megagon Labs
Megagon Labs conducts internships during Winter and Summer terms. Apply for 2026 today!
