Figure 1: amateur photography at NeurIPS
About Me
Figure 2: enjoying some icecream
I’m a Computer Vision Researcher, currently working at Drafted in San Francisco. I graduated from IIT Roorkee with a Bachelor’s in Data Science and AI. During my undergrad I worked across several CV and ML fields, including Diffusion Models, Object Detection, XAI, Geometry Processing and Adversarial ML. I previously led the Data Science Group, IITR’s student-run ML community, as Secretary — building projects, publishing research, and organizing lectures, workshops, and hackathons to drive AI on campus.
Outside of academics, you’ll usually find me on a squash court, out for a run, or scouting for my next ice cream 🍨 or boba fix 🧋.
Latest News
- July 2025 Started my Research Fellowship at MIT!
- July 2025 Our paper ‘DINOHash: Learning Adversarially Robust Perceptual Hashes’ got accepted at the ICML ‘25 CODEML Workshop!
- December 2024 Presented ‘Riemann Sum Optimization for Accurate Integrated Gradients Computation’ at NeurIPS ‘24 IAI Workshop.
Experience
ML Engineer
DraftedJuly 2026 — Present
Building custom diffusion models to architect a variety of floorplans based on customer requirements.
AI Engineer Intern
AlphaSenseDec 2025 — Feb 2025
Worked on Office plugins deploying AI Agents and tools for large finance corporations such as JPMorgan and Goldman Sachs.
Research Fellow — SGI
Massachusetts Institute of TechnologyJul 2025 — Aug 2025
I built new solvers for heat diffusion on discrete meshes in the log domain with Prof. Justin Solomon. I worked on latent-space optimization for DeepSDFs with Paul Kry. I also worked on Grid-Free Fluid Solvers using Implicit Representations with Ishit Mehta and Sina Nabizadeh. This fellowship was actually my first experience with geometry processing but it definitely won’t be the last!
Summer Intern
KLAMay 2025 — Jul 2025
Optimized the YOLO training pipeline for semiconductor defect detection, implemented a deterministic augmentation buffer and a bunch of new augmentations to improve mAP and training speed.
ML Researcher (funded by Zellic)
ProteusJun 2024 — Feb 2024
Worked on robust perceptual hashing, adversarial training, and segmentation for AI-generated content provenance; eventually leading to a paper at the ICML `25 CODEML workshop.
Software Engineering Intern
UberMay 2024 — Jun 2024
Built Go backend services and Kafka-based event streams for data-science experiment notifications.
Publications
Multi-branch trajectory alignment for diffusion distillation: the student maps every teacher timestep at once, recovering the structure sparse supervision throws away.
Adversarially robust perceptual hashing using self-supervised features; open-sourced model and attack tooling.
RiemannOpt: optimizing sample points for Integrated Gradients to reduce computation while preserving attribution quality.
Reproducibility study correcting and analyzing the IDGI framework - experimentally and theoretically.
Projects
LLMs
RAG
Agentic AI
DevRev — AI Agent 007 (Inter-IIT Tech Meet 12.0)
Tooling to generate specialized API calls from natural-language prompts; modified Reflexion and Retriever integration.
XGBoost
Vision Transformers
Computer Vision
Low-Light Image Enhancement
Developed a quantile regression model using XGBoost to create histogram mapping from low-light to high-light images. Achieved improved image quality with vision transformers, increasing mean PSNR on the test set by 3.2.
Web Sockets
Data Structures
CanvArt
Real-time video streaming application using Python sockets with tile-based image compression. Utilized KD Trees and hash tables for matching visually similar images in O(log n) complexity.
Get in Touch
My inbox is always open — singhishree@gmail.com. I don’t check X too often, however, you can reach out to me on LinkedIn. I love connecting with people, so feel free to say hi!