Ved Umrajkar
Final-year BS-MS student in Mathematics and Computing at IIT Roorkee, focused on trustworthy machine learning, computer vision, and applied statistics for real-world systems.
Experience across macro-fee forecasting at JPMorgan Chase, risk analytics on Starknet at Hashstack Finance, and agentic evidence selection research at PavoAI — with papers at ICLR 2025, ICCV 2025, and KDD 2026.
TL;DR of my path so far :)
A timeline of my academic, research, industry, and community milestones.
Education
2021
IIT Roorkee
BS-MS, Mathematics & Computing
Enrolled in the 5-year integrated program. CGPA 8.605. Specialised in ML, Financial Risk, and Statistical Inference.
Swag
2023–25
Data Science Group, IIT Roorkee
Joint Secretary
Mentored students in ML, coordinated campus projects and workshops, and hosted a supervised-learning hackathon on the UCI Gesture Phase Segmentation dataset.
Industry
Aug 2023
Hashstack Finance
Data Scientist
Built a Starknet risk dashboard, boosted anomaly coverage by 15% with a mean-variance optimiser, and shipped automated growth alerts. Aug 2023 – Feb 2024.
Swag
2024
Inter-IIT Tech Meet 13.0
Silver Medal · IIT Bombay
Silver medal building an AI-powered product for Dream11. Also placed 16th in Amazon ML Challenge 2024 (10,000+ candidates) and qualified for Citadel's India Terminal Coding competition.
Research
Jan 2025
Tree Ring Watermarks
ICLR 2025 Workshop Paper
Benchmarked Tree Ring watermark detectability on DDIM vs rectified flow transformers (FLUX.1-dev); diagnosed inversion instability and proposed robust recovery settings for text-to-image safety pipelines.
Swag
Apr 2025
ICLR 2025
Singapore · Conference
Attended the International Conference on Learning Representations. Presented Tree Ring watermark workshop paper at Singapore Expo.
Industry
May 2025
JPMorgan Chase
Software Engineering Intern · Bengaluru
Productionised a macroeconomic-indicator-driven investment management fees forecasting model and built COVID-aware headcount sensitivity prototypes for Asset & Wealth Management. May – Jul 2025.
Research
2025
DAC-LoRA
ICCV 2025 — Workshop on Safe Multimodal AI
Dynamic Adversarial Curriculum for LoRA-based VLM fine-tuning, outperforming vanilla LoRA and single-stage adversarial training on few-shot benchmarks.
Research
Feb 2026
PavoAI
Research Intern · Remote
Co-authored 'Agentic Evidence Selection for Decision-Making under Cognitive Constraints' (KDD 2026 Research Track); formalised task-agnostic evidence selection as constrained submodular optimisation achieving 95.4% of theoretical maximum semantic coverage.