Harshdip Saha
VERIFIEDUndergraduate ML Researcher in Medical Brain Imaging & LLM Safety Alignment
Biographical Summary
Researcher at NexGenLab NSUT & IIT Madras. Core contributor to open-source BrainGlobe. Placed 3rd worldwide in BraTS Lighthouse 2025 (MICCAI 2025). Collaborating with ETH Zurich professors on LLM safety and refusal mechanism dynamics.
Official LinkedIn Record
Public Instagram Record
Extracted Traits & Grounded Evidence
Core Passions (with evidence)
BrainGlobe atlas registration core contributor (15+ merged PRs)
Crosscoder training across fine-tuning checkpoints to track refusal degradation
Author of open-source MCP tools and Agentic autonomous platforms
Observed Hobbies
Daily arXiv deep-dives into LLM alignment and diffusion models
Precision brewing routines noted in workstation captures
Intense intellectual rigor, high agency, first-principles deduction, and relentless curiosity.
Deep work blocks balanced with quiet contemplation, reading, and architectural sketching.
analytical
hyper active
Honest Unknowns (Privacy Contract)
Twofold strictly refrains from inferring sexuality, private romantic expectations, or domestic preferences from social media photos or occupation.
- •Absence of evidence regarding private relationship pace or domestic preferences.
- •Unknown long-term residential preference outside academic/research hubs.
Interactive Personal MCP Console
Reviewers can query Harshdip Saha's Model Context Protocol server directly. Tool invocations execute against verified profile parameters.