Hi! My name is Vibhor Kumar. I'm a computer scientist, amateur neuroscientist, and armchair philosopher of science. I like working at the intersection of the theoretical and applied. My work has been involved in mapping fly brains, catching financial fraud, and generating assets of various types using AI.
I currently work on the Trust & Safety team @Tinder where I combine skills in ML, Data Science, and Backend to catch bad actors and keep Tinder safe. I also advise AI startups on the side.
Feel free to check out my personal website at veezbo.com to learn more!
Repo: https://github.com/veezbo/akkadian_oracle
Live Demo: https://poe.com/AkkadianOracle
A chatbot built using the Poe platform on top of ChatGPT and GPT-4 with Retrieval-Augmented Generation (RAG) with my released Akkadian English corpus
Additional experiments involve pre-trained LLM fine-tuning as well as PEFT (with QLoRA). Notebooks included in the repo.
Research paper to be posted soon
Repo: https://github.com/veezbo/llourney-dev
Collaborators: @dg845 @andrewryanchama
Exploring the feasibility of replacing CLIP with an LLM fine-tuned using a frozen Stable Diffusion model and diffusion training loop.
veezbo/akkadian_oracle
veezbo/akkadian_english_corpus - corresponding HuggingFace dataset
outlines-dev/outlines#287
huggingface/transformers#25807