I'm a Co-Founder of Cerebral Valley, where I focus on search, backend, scraping, and various engineering tasks, while also leading government/defense initiatives, organizing community events, and handling operational responsibilities. I originally joined as Founding Engineer, but proved quite valuable in many other aspects of the business, and was retroactively granted founding equity & title.
For ~2.5 years, I was the first engineering hire on Nima Capital's liquid strategy team (at 19 yo), managing cloud infrastructure, automating trading strategies, building MEV bots, AI engineering for VC workflow and equities analysis, sourcing / diligencing founders, and writing various internal tools for social media and alerting.
LazyTorch: basic PyTorch implemented sequentially (no parallelization) directly in vanilla Python. Complete with Autodiff, Tensors, Loss Functions, Activations, Layers, and full Network abstractions, ready to train on!
eXpect: agentic simulation to predict sentiment and responses from your own X followers for any poll or post. 2nd place in the first x.ai hackathon, featured on their main account here.
SwiftAIv2: an LM fine tuned to write new Taylor Swift songs (improved upon v1).
VLite2: implementing metadata retrieval, fine-grained chunking, and Weaviate's "autocut" feature to avoid selecting K by clustering similar data. This was at one point the fastest available local vector, database powered by usearch, according to benchmarks here.
GPTCacheLite: a semantic caching system for natural language queries, saving 3x compute and LLM inference time for repeated semantically similar queries.
Instant LLM API: boilerplate code to turn any VM into a suite of inference and fine tuning endpoints for any Huggingface LLM, fully compatible with the Llama stack, in a few hundred lines of code.
Figma Translate: tool to convert any Figma file into accurate, working, and well-documented React/Bootstrap code in a dedicated runtime and edit environment.
Word CMD+K: added cursor-style command+k functionality into Microsoft Word.
Mockingbird: an ML tool to automatically DJ a party, generate song recommendations entirely based on crowd feedback, and manage music queuing for establishments. This is my longest ongoing project. I'm trying to find more time to optimize and release this.