I've designed Search & AI infra software deployed on hundreds of millions of devices, which power some of the largest tech companies, unicorns and decacorns, AI research labs, and defense and intelligence organizations worldwide. In short:
- CS/AI researcher.
- Astrophysics dropout.
- Ex Investor: cloud & semiconductors.
- Fluent in English, Russian, & Armenian.
- Building the Unum AI infra project since 2015.
I spent most of the last 20 years writing code. Mostly GPGPU, SIMD Assembly, userspace storage drivers, and networking protocols. I strongly believe, that:
- procedural programming eats object-oriented for breakfast,
- abstractions are mostly evil, they are never zero-cost,
- infrastructure is fun,
- less code is better than more code,
- east-
const
is superior toconst
-west in C/C++, goto
is a blessing, not a curse,- good is the enemy of great β elegant code is better than working code,
- the future of AI is amazing, and it's not LLMs or transformers,
- matrix multiplications are testy, but literally everyone cooks them wrong,
- companies grow faster than people,
- pivots are the plague of Silicon Valley,
- hard things are easy, and easy things are hard,
- talent, ambition, and dedication are much more important than material resources,
- a single developer can beat a trillion-dollar company in almost everything.
If you want to get in touch and contribute - my handle is @ashvardanian on most platforms, including Twitter and LinkedIn.
I've designed and maintain the following libraries, datasets, and AI models:
- StringZilla - faster strings with SIMD/SWAR for C, C++, Rust, Python, and Swift, replacing GlibC and STL
- USearch - single-file search engine for C, C++, Rust, Swift, Go, Java, C#, Python, JS..., replacing FAISS
- UForm - small multimodal pre-trained AI models with SDKs for Python, JS, and Swift, replacing CLIP
- UCall - networking server library for C and Python designed for Linux kernel bypass, replacing FastAPI
- SimSIMD - fast vector-vector math library for C, Python, Rust, and JavaScript, replacing BLAS level 1
Micro repositories:
- USearch-Molecules - 28 billion embeddings - one of the largest datasets in Computational Chemistry
- UStore - multimodal embedded database for C, C++, and Python designed around key-value stores
- SwiftSemanticSearch - example of on-device real-time AI using UForm and USearch on iOS
- cpp-cuda-python-starter-kit - C++ and CUDA starter kit for Python developers
- TinySemVer - Semantic versioning GitHub CI tool that doesn't take 300K lines of JavaScript
- affine-gaps - Less wrong local and global Gotoh sequence alignments in one NumBa Python file
- ParallelReductionsBenchmark - GPGPU benchmarks for SyCL, CUDA, OpenCL, Vulkan, etc.
- BenchmarkingTutorial - C/C++ tutorial for performance-oriented programming using Google Benchmark
- memchr_vs_stringzilla - Rust micro-benchmark comparing StringZilla to the MemChr crate
- usearch-benchmarks - Billion-scale Vector Search benchmarks against FAISS, Weaviate, Qdrant, etc.
- ucsb - parallel benchmarks for ACID persistent key-value stores, like RocksDB
- scaling-democracy - combinatorial voting method implemented in CUDA for H100 GPUs