Happy to share that our workshop paper “Do You Even Know What I Want? Improving LLMs Ability to Detect Hidden User Intent” was accepted into the International Conference on Automated Planning and Scheduling (ICAPS) as a workshop paper.

One well-documented tendency of LLMs is that they often default to answering user queries rather then ask questions about them. With this, a known but often overlooked cause of value misalignment is a human user misspecifying their goals to an agent based on incorrect beliefs about its current state/capabilities resulting in behavior they might not have expected.

Given such things, in this paper we explore a specific human-robot goal misspecification problem: whether LLMs can reason about goal misalignment and identify good questions which resolve it. Though I’ve worked with LLMs in the past, in this project it was fun to use them in more of an AI Safety/Alignment context to see to what extent this was possible. To find out, you’ll of course have to read the paper…but it’s an interesting one, I promise!

If you happen to be around, the paper will be presented as part of the HAXP Workshop in Dublin, Ireland on June 28th .

On behalf of my amazing advisor, I just got back from the ICLR Conference in Rio de Janeiro, Brazil where I had the opportunity to run my first ever workshop (HCAIR: From Human Cognition to AI Reasoning: Models, Methods, and Applications). 

Overall, it was a great experience and I learned so much, including a lot about AI Safety/Alignment which will be the focus of my PhD dissertation. Currently, I have a few papers that are almost ready for submission focused in this area (including one related to LLMs/hidden user intent, so stay tuned for that!), so attending ICLR to learn more and discuss these ideas couldn’t have come at a better time. 

During the workshop, all the speakers were fantastic but a personal highlight for me was getting to introduce Dr. Been Kim, a researcher from Google Deepmind. I’ve known about her work since I began grad school (TCAV, anyone?) so getting to read her papers and then years later see her speak live was a really cool experience. Also, I love chess and her talk focused on methods for better Human-AI Synergy where the use case was using AI to teach superhuman concepts to chess grandmasters. So, its safe to say that I was engaged from the start .

Away from the conference, Brazil was incredible. The beaches/scenery was beautiful, and I got the chance to do something I’ve always wanted to: visit Christ the Redeemer (one of the Seven Wonders of the World). Seeing this was surreal and honestly, I still can’t believe that it happened. Now…just six more wonders left to go!

I just got back from AAAI in Singapore and it did not disappoint .

On the academic side, I had the chance to present both of our accepted papers. Presenting our GSD paper was especially meaningful as it’s one of my favorite projects I’ve worked on in grad school, and my first main conference paper as a first author that I’ve had accepted. Additionally, there was lots of cool research I got to learn about this year, like AI for wildlife conservation, AI in legal and law enforcement contexts, and a particularly thought-provoking talk on the challenges and opportunities of human-centered AI.

The opening reception was also easily one of the coolest I’ve been to because it was held at the zoo, which had manatees, capybaras, etc. and wild monkeys which screamed from the trees as we walked around (there were “do not feed the monkeys” signs everywhere, which definitely added to the experience. Also, the zoo had pandas but evidently they fell asleep before we got a chance to see them).

Beyond the conference, Singapore itself was fascinating. There’s a very rich blend of cultures there (Malay, Indian, Chinese, etc.), the national mascot is the Merlion (half lion, half fish, which is honestly iconic), and there’s lots of interesting rules. For example, no gum, mandatory tray returns after eating (or you get fined), and a strong emphasis on cleanliness and order. It’s also a very rich city which has a very beautiful skyline at night:

While there I explored a lot, including visits to Little India, Chinatown, Arab Street, the Botanic Gardens (a UNESCO World Heritage site), Gardens by the Bay (the light show was fun), and so much more.

The architecture around the city was also very diverse:

And of course the food was amazing. Also, I’m not much of a drinker but the national beer there is Tiger — which again, another ELITE mascot choice.  

Overall, I’d say the conference was a success, the trip was amazing, and I had a great time catching up with everyone.

And though I was sad to leave…Singapore — this isn’t goodbye forever, just for now.

Adding to the excitement is the fact that the conference will be in Singapore this year — marking what will be my first international trip to an Asian country (!!). 

I’ve already completed my poster in preparation (see below) and can’t wait to share it with everyone once there. 

Check out the papers here.

During my 12 weeks there, I contributed to a two-part project focused on improving the SQL generation capabilities of LLM’s using an agentic approach. By the end of the summer this culminated in a lot of learning as well as the demo paper “QueryGym: Step-by-Step Interaction with Relational Databases” (recently accepted into AAAI 2026!! ).

Along the way, I revisited famailier territory (hello planning with LLM’s), dove into learning/implementing several frameworks previously unknown to me (looking at you Pydantic, LangChain, etc.), engaged in a new kind of debugging (solving LLM context window issues, diagnosing SQL generation errors, etc.), and saw how a bunch of things (SQL, OpenAI Gym, etc.) could be brought together into a really unique project. 

Guiding me through all this was a very high caliber, creative team at IBM who really took the time to mentor and challenge me. Being around them showed me how researchers at that level break apart and attack problems, ask questions, and move ideas from inception to full implementation. This was not only interesting to be a part of but showed me all the ways I need to grow as a researcher to hopefully one day play at that level. 

Research aside, IBM filled the summer with great learning opportunities and fun events. One highlight was a talk by comedy writers experimenting with AI in their creative process. On this panel was Joe Toplyn, a former Late Night with David Letterman writer and creator of Witscript, an AI-powered joke-writing assistant, which was fun. Midway through the summer, IBM also hosted an Intern Appreciation Day, treating us to a dinner cruise around the Statue of Liberty. That was the closest I’ve ever been to it, and the views of the New York City skyline at sunset were something special. 

Beyond IBM, I made the most of being in New York. I took time to explore the area and took a number of side trips, including a visit to Rhode Island to see Providence and my personal favorite, Newport. The latter included doing the Cliff Walk, seeing some of the famous mansions in the area, and driving along Ocean Drive (which had some of the most insane pieces of real estate I’ve ever seen). 

To cap it all off, I then made the drive home, crossing a few things off my bucket list along the way:

• Visiting Harpers Ferry & New River Gorge National Park in West Virginia 
• Trekking through Mammoth Cave National Park in Kentucky 

As always, I can’t believe just how quickly time has gone by but what an action-packed summer it was

This paper provides an overview of our ongoing work on intelligent tutoring systems, highlighting how collaboration with domain experts (in our case, entomologists trying to teach people to classify bee species) shapes the development of effective, human-centered learning systems. It also sets the stage for several other research directions we’re currently pursuing…and gave me an excuse to design a few figures (which I always love doing):

Stay tuned for more work we will be publishing in this direction. And of course, check the paper out here if you want to. 

I’m excited to share that today I successfully defended my master’s thesis, Reducing Goal State Divergence with Environment Design!! This milestone feels especially significant because looking back I can’t help but think of everything it took to get here and how much I’ve grown since being in grad school.

Thank you to everyone who helped me along the way – I’m extremely grateful for your support!

If you’d like to read my full thesis, you can check it out here.

Check my presentation slides out here.

It was great to be back at AAAI this year, especially with the conference being held in Philadelphia, PA. While there, I had the chance to present our workshop paper, Creating PDDL Models from JavaScript Using LLMs: Preliminary Results, reconnect with a bunch of people from the NRL and broader research community as a whole, and of course learned about all sorts of new research!

Two highlights for me were a survey paper — A Survey for Leveraging Large Language Models to Construct Automated Planning Models — which provided an excellent overview of how LLMs are being used to build automated planning models and getting to see Dr. George Konidaris from Brown University speak (who hands down gave my favorite presentation of the entire conference). His lab focuses on the fundamental computational processes that generate intelligence, and how they can be used to design generally-intelligent robots. I found his entire presentation extremely interesting and have since been learning more about the work he and his lab does.

Away from the conference, I also carved out a bit of time to explore the city. I visited the Philadelphia Museum of Art + Rocky Statue, Ben Franklin’s Grave — which felt fitting, since lately I’ve been working my way through The Autobiography of Ben Franklin (It’s so good…I highly recommend it) — and of course stopped by the Liberty Bell, among a few other places.

Overall, it was a fantastic experience, both personally and professionally, and a great reminder of how energizing it can be to spend time in person with colleagues and friends in the field.

Exciting news—my workshop paper, “Creating PDDL Models from JavaScript Using LLMs: Preliminary Results,” has been accepted into AAAI 2025! This work stems from research I did during my time at the Naval Research Lab this past summer. I’m thrilled to be heading to Philadelphia to present it and while there to track down a Philly cheesesteak (or two).

In AI planning, automated planners require formal domain models to generate correct plans. But creating these models isn’t always straightforward—as many know, writing them can be time-consuming and often requires specialized domain knowledge.

In this work, we tackle this challenge by creating a two-stage pipeline that combines natural language and operational models to generate these planning models. Specifically, we convert JavaScript code from an existing robotic controller (in this case, corresponding to Minecraft actions) into PDDL, not only extracting a valid planning model but also testing the the limits of what large language models (LLMs) can do. Based on our findings, we highlight a few techniques and other research directions we feel could enhance the effectiveness of LLMs in future AI planning endeavors.

Check out the paper here.

BIG NEWS – I’ve been selected to intern this summer at the IBM Watson Research Center in Yorktown Heights, NY! I am ecstatic for this opportunity and can’t wait to spend the summer pushing the boundaries of what’s possible in AI.

Stay tuned for more updates on my journey at IBM—this summer’s going to be a good one!