Nandan Sarkar
PhD Student at NYU Courant · Yale ’26
Hey, I’m Nandan! I’m a PhD student in Computer Science at NYU Courant, where I work with Professors Eunsol Choi and Pavel Izmailov. I’m grateful to be funded by the National Science Foundation Graduate Research Fellowship (NSF GRFP). I’m interested in natural language processing and machine learning; broadly, my research interests include designing efficient model architectures, understanding the capabilities and limitations of current systems, and developing models with more human-like generalization and reasoning.
I graduated cum laude from Yale University in 2026 with Distinction in both my majors: Computer Science (B.S.) and Applied Mathematics (B.S.).
At Yale, I spent three amazing years working with Professor Arman Cohan. I was also fortunate to spend two summers in person and two years working remotely at MIT CSAIL with Professor Yoon Kim.
I’ve had the opportunity to learn from incredible mentors throughout my journey, including Abbas Zeitoun, Lucas Torroba-Hennigen, Shruti Singh, and Zhaobo K. Zheng. Be sure to check out their amazing work!
Loading…
Each model gets 150 moves and 20 minutes. Click a model to see what it solved!
How this works
I open this page in a browser, show the model a screenshot and a short set of instructionsYou are solving a jigsaw puzzle on a web page. You see it as a 1280x800 screenshot, and screenshot pixels are exactly the coordinates the drag tool uses. The goal: a 6x6 photo puzzle of 36 pieces. Its empty spot is the wooden frame on the right side of the page, which shows a faint ghost of the finished photo to guide you. The photo area is 240x240 pixels with its top-left corner at (x, y). It's a grid of 40x40 cells, so the cell in column c and row r (both counted 0-5 from the top-left) has its center at (x+20+40c, y+20+40r). The pieces: the 36 loose pieces sit in rows in the blank margins at the left and right edges of the page. All of them are already the right way up; you never need to rotate anything. Each piece has a 40x40 square body, plus tabs that stick out up to about 12 pixels on some sides. Pieces from the edges of the photo have a flat side, and the four corner pieces have two. How dragging works: drag(from_x, from_y, to_x, to_y) presses the mouse at the first point and releases it at the second. The piece moves by exactly that distance, so press on the center of the piece's square body (not on a tab) and release on the center of the cell where it belongs. Snapping: a piece locks into place if you drop its center within 16 pixels of its correct cell center, horizontally and vertically. A locked piece loses its drop shadow and can't be moved again. A piece dropped anywhere else just stays there and may cover other pieces or the ghost image, so don't drop pieces in the frame unless you're fairly confident. Moving a misplaced piece costs another move. Budget: you have 150 drags and 20 minutes in total. After each turn you get a new screenshot, the number of pieces placed, and the moves left. You can make several drags per turn. Keep going until all 36 pieces are placed., and let it play by sending mouse drags. After each turn it gets a new screenshot and its piece count, and it can look back over the whole game so far. The run ends when the puzzle is solved, when the model runs out of moves or time, or when it goes two minutes without moving a piece. Every model gets the same instructions and starting layout, the pieces hold still, and the clock keeps running while the model thinks.
Loading…
