Serve AI (YC S24, NFX, GC) San Francisco, CA
Founding AI Engineer Intern
-
Theoretical research, quantitative trading, and startups.
Revyl (YC F24) San Francisco, CA
Founding AI Engineer Intern
-
MIT PRIMES CrowdMath Cambridge, MA
Researcher
-
- Open-source research on finitely generated rational semirings focusing on monoids, generators, and homomorphisms
Harvard Medical School Boston, MA
ML Research Intern, Dr. Kiran Agarwal-Harding
-
- Fine-tuned Curia FM for pelvic/acetabular fracture detection
MLiNS Lab at Michigan Medicine Ann Arbor, MI
ML Research Intern, Dr. Todd Hollon
-
- Compiled largest open access figure/caption pair dataset from 25K+ neurology papers (NeurIPS 2026 submission)
- Enhanced CLIP/SigLIP for MRI analysis using structured itemized descriptions and FLAIR to align text with localized image regions; fine-tuned zero-shot segmentation in ViTs to improve anatomical delineation accuracy
H.O.P.E. Lab at University of Michigan Dentistry Ann Arbor, MI
Senior Mentor
-
Oakland University Summer Mathematics Institute Rochester, MI
Program Participant
-
- 1 of 14 accepted for graduate theory classes
University of Michigan College of Engineering Ann Arbor, MI
Research Intern, Dr. Seth Pettie
-
- Designed performance benchmarks against existing data sketches for the SymmetricPoissonTower sketch which summarizes large amounts of turnstile streaming data (f -moments) in one pass using limited/sublinear memory
- Exploited harmonic structure to eliminate explicit sampling achieving exponential (10-100x) memory savings and reducing space to O(log² n) over L0/L2 methods, validating it as the universal sketch for redundant data retrieval
- Sampled edge weights to nondeterministically achieve 2-point distribution (modeled after the Fisher distribution) that assigns weights 0.41 and 4.75 with probabilities 44% and 56%, respectively; arXiv:2507.23105 preprint
H.O.P.E. Lab at University of Michigan Dentistry Ann Arbor, MI
Lab Intern, Dr. Alexandre DaSilva
-
- 1 of 3 accepted to work on craniofacial pain therapeutics
University of Waterloo Faculty of Mathematics Waterloo, ON
Research Intern, Dr. Mohammad Kohandel
-
- Extended symbolic regression tools (AI Feynman and UPINNs) to dimensionally reduced physical equations, correctly recovering target equations/terms with 90-99% fewer data points compared to the dimensional expressions
- Accelerated convergence to guarantee the recovery of an algebraic equation or hidden differential term in one attempt improving accuracy to 100% and reducing runtime by 85-90%, building a robust framework to deal with noisy data
- Presented at 2025 International Science & Engineering Fair and Michigan Science & Engineering Fair in Mathematics; arXiv:2411.15919 preprint and submitted to Applied Mathematical Modelling in ScienceDirect (5.1 Impact Factor)
Harvard Medical School & MIT Cognitive Sciences Harvard, MA
ML Research Intern, Dr. Carlos Ponce & Dr. Suayb Arslan
-
- Developed a pipeline that translates EEG signals into visual representations by conditioning a diffusion model (DreamDiffusion) with EEG-derived prompts processed through transformer-based encoders (Stable Diffusion)
- Integrated functionality to convert EEG sleep signals into images on Google Colab codebase achieving a loss of 0.000243, eliminating specific file/repository dependencies and providing the original training data file to users
- Presented at 2024 Science Fair of Metro Detroit in Robotics & Intelligent Machines; arXiv:2407.02673 preprint