about/
Hi, I'm Vishesh (he/him) :)
I'm a junior in CS at the University of Cincinnati (graduating April 2027).
Student researcher in the CincyNLP Lab, advised by Dr. Tianyu Jiang. Currently working on AI Safety and Alignment, aiming to study the causal relationship and emergence of these behaviours with respect to the models internals. I like building things which provide some value to other people. Startups and VC fascinate me, big value creating and economy driving machines, which is partly why I recently served as a Partner at Bearcat Ventures, UC's student-led ~$1M fund.
Love hackathons to build cool stuff and meet even cooler people. Fall'25, I was hacking at HackMIT, Cloudflare, and Calhacks, usually around agents / MCP / whatever sounds fun. If you want to chat research, startups, or random ideas: anandvh@mail.uc.edu.
news/
2026-05: Accepted to YC's Summer 2026 Startup School.
2026-04: "Rhetorical Questions in LLM Representations" accepted to ACL Main.
2025-11: 2nd place, VCIC (Venture Capital Investment Competition) Regionals at CMU.
2025-10: Cal Hacks, 3rd place, Snap AR track (memARy).
2025-08: Joined CincyNLP Lab.
2025-06: Startup Weekend, winner, Edge AI track.
2024-05: TA and Sophomore of the Year (UC).
research/
Got pulled into this field through Anthropic's Toy Models of Superposition, something about that framing of a model's internal geometry just clicked. Went deep on Neel Nanda's work after that, built a small transformer from scratch to get my hands dirty, and it all snowballed from there.
What I keep coming back to is the gap between what a model outputs and what's actually happening in its weights and activations. I care a lot about safety and alignment: what is this model actually doing, and can we catch it before it does something bad? I think the answer lives somewhere in the internals, and interpretability is how we get there.
- Louie Hong Yao, Vishesh Anand, Yuan Zhuang, and Tianyu Jiang.Rhetorical Questions in LLM Representations: A Linear Probing Study.Accepted to the main conference of the 64th Annual Meeting of the Association for Computational Linguistics (ACL 2026). Arxiv.
projects/
Selected builds. More on GitHub and Devpost.
Kite, "Cursor for GitHub", SoarAI Labs, OpenAI OSS Hackathon (Aug 2025 to present).
Agentic CLI and dashboard for natural language git work, PRs, commits, branches, merge conflicts. MCP tooling, RAG over commits. github, site, devpost.
memARy, Cal Hacks 2025, 3rd place, Snap AR track.
Wearable-ish memory, vision and audio to structured search (Reka, ChromaDB RAG, MCP on Render). devpost.
XStream, MCPLite Hackathon at Cloudflare, 2025.
TypeScript MCP server on Cloudflare Workers, post to X with OAuth, D1, Workers AI images, built in about two hours.
Just-Hurry!, Hack Harvard 2024.
Hurricane impact zones and volunteer coordination in under 36 hours. Next.js, Flask, MongoDB. devpost.
experience/
ML Intern (CV), Kinetic Vision, Jan 2026 to May 2026.
Research Assistant, CincyNLP Lab, Dr. Tianyu Jiang, Aug 2025 to present.
Linear probing and causal steering experiments to understand how rhetorical intent is geometrically encoded across LLM layers. The bigger picture is mechanistic interpretability for safety: figuring out whether alignment relevant behaviors leave a legible trace in activations, and if so, whether we can intervene on them before deployment.
General Partner, Bearcat Ventures ($1M Student Fund), Jan 2025 to present.
Seed and pre-seed diligence, helped shape $25K SAFEs, led a $100K Series A for a TBI headset company, deployed ~$200k in total. Mentored a partner group of 5 analysts. Built scraping and LinkedIn tools that roughly doubled our deal flow pipeline as the director of deal flow.
AI Solutions Co-op, Kinetic Vision, Apr to Jul 2024.
Voice kiosk "Lucy" for UC's 1819 Innovation Hub, shipped end to end. Also YOLOv5 for product ID and B2B testing platform work. UC write-up.
Software Engineering Co-op, Kinetic Vision, Aug to Dec 2023.
Production volunteer portal (TypeScript, Keystone, React). Internal Resource Mapper, in production currently.
Research Assistant, Bioinformatics, Dr. Danny Wu, Jan 2024 to Jan 2025.
Full-stack dashboard for Cincinnati Children's HIRC (200+ researchers), Flask, SQL tuning, a lot of tickets closed.
tidbits/
- Four hackathons in one semester, placed or won most of them. Apparently I function very well on no sleep. (Do not try this at home.)
- When I'm not parked in front of a screen, rare, but I try, I like the gym.