- 🎓 Final-year M.Tech (Research) at IISc, CDS — working with VAL / VCL.
- 🔭 Research interests: sketch-guided localization, hand–object interaction generation, 3D scene representations (NeRF, 3D Gaussians), and grounded detection.
- ⚙️ Love building research-grade systems: custom PyTorch ops, DDP/NCCL, SLURM on DGX, Dockerized reproducible setups.
- 🌱 Currently exploring: triplane-guided HOI generation, sketch-conditioned GroundingDINO, and agentic ML systems for real-world impact.
- Sketch-conditioned GroundingDINO — extended grounding with a SketchEncoder for retrieval-aligned detection.
- HOI-Diffusion — triplane intermediate representation for coherent hand–object trajectories in 3D.
- 3D Gaussian Workflows — experiments on generalizable splats & NeRF pipelines with robust camera handling.
- Award — Co-author on IAPR Best Paper (CVIP 2023) for vision-based fire detection & classification.
I enjoy turning messy research ideas into clean, reproducible repos with good docs, configs, and ablations.
- 🧩 Better open-set / sketch-guided localization for real-world categories.
- 👐 Consistent HOI generation with physically plausible contacts.
- 🧱 Infra: multi-GPU training (DDP/NCCL), data pipelines, and robust loaders for large mixed-modality datasets.
I refactor dataloaders more than I refactor life. Also: dark mode forever.