https://ashvardanian.com/ Recent content on Ash's Blog https://ashvardanian.com/ashvardanian.com.jpg https://ashvardanian.com/ashvardanian.com.jpg Hugo -- 0.139.2 en Fri, 20 Mar 2026 00:00:00 +0000 https://ashvardanian.com/posts/numkong/ Fri, 20 Mar 2026 00:00:00 +0000 https://ashvardanian.com/posts/numkong/ Around 2&#39;000 SIMD kernels for mixed-precision BLAS-like numerics — dot products, batched GEMMs, distances, geospatial, ColBERT MaxSim, and mesh alignment — from Float6 to Float118, leveraging RISC-V, Intel AMX, Arm SME, and WebAssembly Relaxed SIMD, in 7 languages and 5 MB. https://ashvardanian.com/posts/search-utf8/ Mon, 15 Dec 2025 00:00:00 +0000 https://ashvardanian.com/posts/search-utf8/ ICU gets Unicode right and pays for it. This post shows a different approach: fold-safe windows, SIMD probes, and verifiers for fast UTF‑8 search. https://ashvardanian.com/posts/chacha-vs-aes-2025/ Fri, 07 Nov 2025 00:00:00 +0000 https://ashvardanian.com/posts/chacha-vs-aes-2025/ AES-256-GCM now beats ChaCha20-Poly1305 by up to 3x on every modern CPU with hardware acceleration, reversing the 2015 mobile performance advice. https://ashvardanian.com/posts/scaling-elections/ Tue, 28 Oct 2025 00:00:00 +0000 https://ashvardanian.com/posts/scaling-elections/ Taking the computationally expensive Schulze voting method from theory to practice on GPUs — exploring parallel algorithms, hardware optimizations, and why Mojo surprised me. https://ashvardanian.com/posts/aws-graviton-checksums-on-neon-vs-sve/ Mon, 06 Oct 2025 00:00:00 +0000 https://ashvardanian.com/posts/aws-graviton-checksums-on-neon-vs-sve/ SVE2 AES on AWS Graviton 4 delivers 2x faster string hashing than NEON, but SVE&#39;s variable-length promise fell flat: Graviton 4 regressed to 128-bit vectors. https://ashvardanian.com/posts/llm-epicycles/ Thu, 02 Oct 2025 00:00:00 +0000 https://ashvardanian.com/posts/llm-epicycles/ Like Ptolemaic astronomers stacking circles to predict planetary motion, we stack transformer layers to approximate intelligence—accurate yet awaiting our Kepler moment. https://ashvardanian.com/posts/image-processing-with-strings/ Sat, 20 Sep 2025 00:00:00 +0000 https://ashvardanian.com/posts/image-processing-with-strings/ StringZilla&#39;s SIMD-optimized Look-Up Tables beat OpenCV by 4x on image processing, proving string manipulation techniques excel beyond text. https://ashvardanian.com/posts/stringwars-on-gpus/ Mon, 15 Sep 2025 00:00:00 +0000 https://ashvardanian.com/posts/stringwars-on-gpus/ StringZilla v4 brings CUDA acceleration for string processing: 109x faster than Nvidia&#39;s CuDF on edit distances, plus 52-bit MinHash fingerprinting and AES-based hashing. https://ashvardanian.com/posts/beyond-openmp-in-cpp-rust/ Tue, 20 May 2025 00:00:00 +0000 https://ashvardanian.com/posts/beyond-openmp-in-cpp-rust/ Most C++ and Rust thread pools run 10x slower than OpenMP on fork-join workloads. Fork Union closes the gap to 20% with 300 lines, dodging mutexes and CAS. https://ashvardanian.com/posts/less-wrong-cuda-hello-world/ Sat, 05 Apr 2025 00:00:00 +0000 https://ashvardanian.com/posts/less-wrong-cuda-hello-world/ Moving beyond triple-bracket kernel launches to production-ready CUDA: proper error handling, CUDA Driver API, cooperative groups, and inline PTX for robust GPU code. https://ashvardanian.com/posts/longest-ptx-instruction/ Wed, 05 Feb 2025 00:00:00 +0000 https://ashvardanian.com/posts/longest-ptx-instruction/ Exploring the 79-character mma.sp::ordered_metadata instruction for Tensor Cores on Hopper H100s, plus insights on PTX, SASS, and the evolving complexity of GPU ISAs. https://ashvardanian.com/posts/cpu-ports/ Sun, 19 Jan 2025 00:00:00 +0000 https://ashvardanian.com/posts/cpu-ports/ By exploiting CPU port differences between Intel and AMD, interleaving FMA with addition instructions boosts AVX-512 reduction kernels from 211 GB/s to 330 GB/s per core. https://ashvardanian.com/posts/parsing-json-with-allocators-cpp/ Tue, 07 Jan 2025 00:00:00 +0000 https://ashvardanian.com/posts/parsing-json-with-allocators-cpp/ Custom allocators can dramatically reduce JSON parsing overhead, but singleton dependencies lurk everywhere—from malloc to std::isspace—limiting multi-threaded gains. https://ashvardanian.com/posts/splitting-strings-cpp/ Thu, 02 Jan 2025 00:00:00 +0000 https://ashvardanian.com/posts/splitting-strings-cpp/ A 16-year-old StackOverflow question on string splitting gets a modern answer: SIMD-accelerated tokenization that&#39;s 10x faster than STL and cleaner than ranges. https://ashvardanian.com/posts/next-31-years-of-unum/ Tue, 26 Nov 2024 00:00:00 +0000 https://ashvardanian.com/posts/next-31-years-of-unum/ Nine years into a 40-year commitment to AI infrastructure. Reflecting on mistakes, open-source wins, and why the real work is just beginning. https://ashvardanian.com/posts/understanding-simd-complexity/ Mon, 25 Nov 2024 00:00:00 +0000 https://ashvardanian.com/posts/understanding-simd-complexity/ Why SIMD programming is harder than it looks: exploring cosine similarity optimization across x86 and Arm architectures, from basic AVX2 to cutting-edge SVE2 instructions. https://ashvardanian.com/posts/simd-set-intersections-sve2-avx512/ Mon, 16 Sep 2024 00:00:00 +0000 https://ashvardanian.com/posts/simd-set-intersections-sve2-avx512/ Achieving 5x faster set intersections on AWS Graviton 4 using Arm&#39;s SVE2 and Intel&#39;s AVX-512 with specialized instructions like HISTCNT, MATCH, and VP2INTERSECT. https://ashvardanian.com/posts/discount-on-keyword-arguments-in-python/ Sun, 08 Sep 2024 00:00:00 +0000 https://ashvardanian.com/posts/discount-on-keyword-arguments-in-python/ Keyword arguments in Python aren&#39;t free. Switching from PyArg_ParseTupleAndKeywords to METH_FASTCALL and manual parsing yields a 35% speedup in function calls. https://ashvardanian.com/posts/numpy-vs-blas-costs/ Tue, 12 Mar 2024 00:00:00 +0000 https://ashvardanian.com/posts/numpy-vs-blas-costs/ NumPy&#39;s binding overhead wastes up to 90% of BLAS throughput on dot products; SimSIMD recovers that performance with optimized Python interfaces. https://ashvardanian.com/posts/painful-strings/ Mon, 12 Feb 2024 00:00:00 +0000 https://ashvardanian.com/posts/painful-strings/ C++ STL strings bring 20,000+ lines of code to every compilation, yet remain slower than LibC and error-prone. StringZilla offers a faster, more intuitive alternative. https://ashvardanian.com/posts/usearch-molecules/ Mon, 20 Nov 2023 00:00:00 +0000 https://ashvardanian.com/posts/usearch-molecules/ Indexing 28 billion chemical embeddings from 7 billion molecules using optimized Jaccard distance kernels, achieving 99% recall at 3,700 QPS on AWS Open Data. https://ashvardanian.com/posts/porting-cpp-library-to-ten-languages/ Thu, 09 Nov 2023 00:00:00 +0000 https://ashvardanian.com/posts/porting-cpp-library-to-ten-languages/ Binding USearch to 10 languages reveals the friction in each ecosystem—from Python&#39;s simplicity to Java&#39;s complexity—and why C++ still rules performance-critical code. https://ashvardanian.com/posts/python-c-assembly-comparison/ Mon, 30 Oct 2023 00:00:00 +0000 https://ashvardanian.com/posts/python-c-assembly-comparison/ From pure Python to AVX-512 assembly, optimizing cosine similarity reveals a 2,500x speedup through SIMD, FP16, and VNNI instructions on modern Intel CPUs. https://ashvardanian.com/posts/gcc-12-vs-avx512fp16/ Mon, 23 Oct 2023 00:00:00 +0000 https://ashvardanian.com/posts/gcc-12-vs-avx512fp16/ Hand-written AVX-512FP16 SIMD code achieves 119x speedup over GCC&#39;s auto-vectorization for Jensen-Shannon divergence, using reciprocal square roots and FMA instructions. https://ashvardanian.com/posts/javascript-ai-vector-search/ Sat, 21 Oct 2023 00:00:00 +0000 https://ashvardanian.com/posts/javascript-ai-vector-search/ Supercharging JavaScript&#39;s TypedArray with C bindings and SIMD achieves 10x speedup over native Arrays for AI vector operations in Node.js. https://ashvardanian.com/posts/pybind11-cpython-tutorial/ Tue, 10 Oct 2023 00:00:00 +0000 https://ashvardanian.com/posts/pybind11-cpython-tutorial/ Switching from PyBind11 to direct CPython C API bindings reduced StringZilla&#39;s call latency by 5x, proving that understanding CPython internals pays dividends. https://ashvardanian.com/posts/simsimd-faster-scipy/ Sat, 07 Oct 2023 00:00:00 +0000 https://ashvardanian.com/posts/simsimd-faster-scipy/ AVX-512 and Arm SVE bring masked loads and native f16 math to vector distances. SimSIMD exploits both, running up to 200x faster than SciPy on modern CPUs. https://ashvardanian.com/posts/searching-stable-marriages/ Tue, 18 Jul 2023 00:00:00 +0000 https://ashvardanian.com/posts/searching-stable-marriages/ Replacing billion-entry preference lists with vector search indexes makes the Nobel Prize-winning Stable Marriage algorithm scale to real-world dating and database joins. https://ashvardanian.com/posts/stringzilla/ Mon, 10 Jul 2023 00:00:00 +0000 https://ashvardanian.com/posts/stringzilla/ StringZilla uses SIMD tricks to hit 16 GB/s substring search, beating standard libraries by 5-10x and parsing multi-terabyte files Python couldn&#39;t handle. https://ashvardanian.com/posts/abusing-vector-search/ Tue, 09 May 2023 00:00:00 +0000 https://ashvardanian.com/posts/abusing-vector-search/ Vector search isn&#39;t just for AI embeddings. Use HNSW for geo-spatial queries, stock covariance, chess positions, text tokens, and multi-modal recommendations with custom metrics. https://ashvardanian.com/posts/count-unique-strings/ Tue, 09 May 2023 00:00:00 +0000 https://ashvardanian.com/posts/count-unique-strings/ From junior to expert: four C++ solutions to count unique strings reveal 30x performance gaps. Proficiency dramatically impacts even trivial single-threaded code. https://ashvardanian.com/posts/my-sona/ Fri, 29 Apr 2022 00:00:00 +0000 https://ashvardanian.com/posts/my-sona/ A love story cut short — remembering Sona, who taught me what it means to truly care for others, build with purpose, and find someone who shares your frequency. https://ashvardanian.com/posts/google-benchmark/ Fri, 04 Mar 2022 00:00:00 +0000 https://ashvardanian.com/posts/google-benchmark/ Write faster C++ with Google Benchmark: measure everything from math micro-ops to sorting algorithms, and discover 60x speedups with simple compiler tricks. https://ashvardanian.com/posts/ddr4-bandwidth/ Sat, 29 Jan 2022 00:00:00 +0000 https://ashvardanian.com/posts/ddr4-bandwidth/ Why reaching DDR4&#39;s theoretical bandwidth is nearly impossible. CPU parallel reductions struggle to hit 60% saturation while GPU solutions easily exceed 79%. https://ashvardanian.com/posts/cuda-parallel-reductions/ Fri, 28 Jan 2022 00:00:00 +0000 https://ashvardanian.com/posts/cuda-parallel-reductions/ GPU code beats optimized CPU parallel reductions by 10x, reaching 879 GB/s. CUB achieves 94% bandwidth saturation while CPU barely hits 60%. https://ashvardanian.com/posts/apple-m1/ Tue, 21 Dec 2021 00:00:00 +0000 https://ashvardanian.com/posts/apple-m1/ M1 Max with DDR5 challenges 64-core server workloads in hash-table benchmarks. Memory bandwidth matters more than core count for many real-world tasks. https://ashvardanian.com/posts/server-supercycle/ Tue, 07 Dec 2021 00:00:00 +0000 https://ashvardanian.com/posts/server-supercycle/ DDR5, PCIe Gen5, CXL, and new CPUs from Intel, AMD, and NVIDIA converge in 2022. The biggest datacenter hardware refresh in a decade. https://ashvardanian.com/posts/simd-popularity/ Sun, 21 Nov 2021 00:00:00 +0000 https://ashvardanian.com/posts/simd-popularity/ Analysis of 2,000+ Linux and macOS binaries reveals less than 1% of instructions use SIMD, despite CPUs dedicating massive die area to vector processing. https://ashvardanian.com/posts/artsakh/ Fri, 02 Oct 2020 00:00:00 +0000 https://ashvardanian.com/posts/artsakh/ A data-driven argument for why Artsakh must be independent — examining historical claims, demographic evidence, and the indisputable pattern of ethnic violence. https://ashvardanian.com/posts/turkey/ Thu, 01 Oct 2020 00:00:00 +0000 https://ashvardanian.com/posts/turkey/ Seven stories tracing Turkey&#39;s path from secular reforms to neo-Ottoman ambitions — examining crimes unpunished, allies betrayed, and the dangerous pattern repeating today. https://ashvardanian.com/posts/aggressor/ Sun, 27 Sep 2020 00:00:00 +0000 https://ashvardanian.com/posts/aggressor/ Three countries, vastly different resources and freedoms. The numbers reveal who&#39;s really threatening peace in the region. https://ashvardanian.com/posts/armenia/ Sat, 01 Aug 2020 00:00:00 +0000 https://ashvardanian.com/posts/armenia/ Why Armenia is an emerging tech hub worth your attention — from startup tax incentives to ancient history, discover what makes this small nation punch above its weight. https://ashvardanian.com/posts/covid19/ Sun, 22 Mar 2020 00:00:00 +0000 https://ashvardanian.com/posts/covid19/ The pandemic revealed our weaknesses, but also unprecedented opportunities — from personal growth to financial markets, here&#39;s why optimism isn&#39;t naive. https://ashvardanian.com/posts/building-ai-safely/ Fri, 06 Jul 2018 00:00:00 +0000 https://ashvardanian.com/posts/building-ai-safely/ A 2018 conversation about AI safety, infrastructure, and the real dangers of weak AI in the hands of organizations optimizing for power over people. https://ashvardanian.com/posts/whats-wrong-with-wwdc-2016/ Tue, 14 Jun 2016 00:00:00 +0000 https://ashvardanian.com/posts/whats-wrong-with-wwdc-2016/ An iOS developer&#39;s critique of Apple&#39;s 2016 keynote — when emojis and animations overshadowed the innovations developers actually needed. https://ashvardanian.com/hackernews/ Mon, 01 Jan 0001 00:00:00 +0000 https://ashvardanian.com/hackernews/ <p>I often post on Reddit and HackerNews, with the latter being a surprisingly well-balanced platform for technical discussions with less personal bias. Here are some of my favorite publications that made headlines on HackerNews:</p> <ol> <li><a href="https://news.ycombinator.com/item?id=35042316">Up to 100x Faster FastAPI with simdjson and io_uring on Linux 5.19</a>.</li> <li><a href="https://news.ycombinator.com/item?id=34970045">Beating OpenAI CLIP with 100x less data and compute</a>.</li> <li><a href="https://news.ycombinator.com/item?id=43727743">Less Slow C++</a>.</li> <li><a href="https://news.ycombinator.com/item?id=46276826">Full Unicode Search at 50× ICU Speed with AVX‑512</a>.</li> <li><a href="https://news.ycombinator.com/item?id=45402820">Beyond OpenMP in C++ and Rust: Taskflow, Rayon, Fork Union</a></li> <li><a href="https://news.ycombinator.com/item?id=45304807">Processing Strings 109x Faster Than Nvidia on H100</a>.</li> <li><a href="https://news.ycombinator.com/item?id=42237938">Understanding SIMD: Infinite complexity of trivial problems</a></li> <li><a href="https://news.ycombinator.com/item?id=35887983">Abusing vector search for texts, maps, and chess</a>.</li> <li><a href="https://news.ycombinator.com/item?id=29670624">Apple to Apple Comparison: M1 Max vs. Intel</a>.</li> <li><a href="https://news.ycombinator.com/item?id=29954447">Server Hardware Super-Cycle 2022</a>.</li> <li><a href="https://news.ycombinator.com/item?id=30178764">Failing to reach DDR4 bandwidth</a>.</li> <li><a href="https://news.ycombinator.com/item?id=38684461">Python, C, Assembly – Faster Cosine Similarity</a>.</li> <li><a href="https://news.ycombinator.com/item?id=36772545">Combinatorial Stable Marriages for DBMS Semantic Joins</a>.</li> <li><a href="https://news.ycombinator.com/item?id=37273963">Faking SIMD to Search and Sort Strings 5x Faster</a>.</li> <li><a href="https://news.ycombinator.com/item?id=39111114">SimSIMD v3.6.7: Hardware-accelerated similarity metrics and distance functions</a>.</li> <li><a href="https://news.ycombinator.com/item?id=37304306">StringZilla v2: Fastest string sort, search, split, and shuffle using SIMD</a>.</li> </ol> https://ashvardanian.com/software/ Mon, 01 Jan 0001 00:00:00 +0000 https://ashvardanian.com/software/ <p>All of my software is hosted on GitHub, mostly under the <a href="https://opensource.org/licenses/Apache-2.0">Apache-2.0</a> permissive license. Free for commercial and non-commercial use, modification, and distribution.</p> <h2 id="major-projects">Major Projects</h2> <ul> <li><strong><a href="https://github.com/unum-cloud/USearch">USearch</a></strong> - a universal search engine powering many databases, AI labs, and experiments in Natural Sciences. Compact C++ core with 10+ language bindings — 10–100× faster than Meta FAISS for vector search and far beyond Apache Lucene.</li> <li><strong><a href="https://github.com/ashvardanian/StringZilla">StringZilla</a></strong> - SIMD, SWAR, and CUDA-accelerated string algorithms for search, matching, hashing, and sorting at Web Scale and Bioinformatics scale. Hundreds of hand-tuned kernels with manual multi-versioning, exposed to C, C++, Rust, Python, Swift, and JavaScript, up to 10× faster on CPUs and 100× faster on GPUs.</li> <li><strong><a href="https://github.com/ashvardanian/SimSIMD">SimSIMD</a></strong> - an extensive collection of mixed-precision vector math kernels for C, Python, Rust, and JavaScript. Designed for linear algebra, scientific computing, statistics, information retrieval, and image processing, delivering consistent SIMD speedups over BLAS and NumPy on both x86 and ARM architectures.</li> <li><strong><a href="https://github.com/unum-cloud/UCall">UCall</a></strong> - a kernel-bypass web server backend for C and Python built on io_uring. Achieves 70× higher throughput and 50× lower latency than FastAPI for real-time workloads, including serving compact AI models.</li> <li><strong><a href="https://github.com/unum-cloud/UForm">UForm</a></strong> - tiny multimodal AI models with state-of-the-art parameter and data efficiency. Compatible with Python, JS, and Swift, serving as a lightweight alternative to OpenAI CLIP for on-device and server inference.</li> <li><strong><a href="https://github.com/ashvardanian/ForkUnion">ForkUnion</a></strong> - ultra-low-latency parallelism library for Rust and C++. Avoids allocations, mutexes, and even Compare-And-Swap atomics — achieving up to 10× speedups over Rayon and TaskFlow.</li> </ul> <p>Some of those are used in open-source databases, like <a href="https://github.com/ClickHouse/ClickHouse">ClickHouse</a>, <a href="https://github.com/duckdb/duckdb">DuckDB</a>, <a href="https://github.com/pingcap/tidb">TiDB</a>, <a href="https://github.com/scylladb/scylladb">ScyllaDB</a>, <a href="https://github.com/yugabyte/yugabyte-db">yugabyteDB</a>, <a href="https://github.com/dragonflydb/dragonfly">DragonflyDB</a>, <a href="https://github.com/memgraph">MemGraph</a>, <a href="https://github.com/vdaas/vald">Vald</a>, <a href="https://github.com/tursodatabase/turso">Turso</a>, LLM toolchains, like <a href="https://github.com/langchain-ai/langchain">LangChain</a>, <a href="https://github.com/run-llama/semtools">LlamaIndex</a>, <a href="https://github.com/microsoft/semantic-kernel">Microsoft SemanticKernel</a>, <a href="https://github.com/nomic-ai/gpt4all">Nomic AI GPT4All</a>, <a href="https://github.com/deta/surf">Surf</a>, and many other less &ldquo;open&rdquo; systems, such as backend infrastructure of major AI labs, government intelligence agencies, Hyper-scale cloud companies, Fortune 500, iOS and Android apps with 100M-1B MAU.</p> https://ashvardanian.com/talks/ Mon, 01 Jan 0001 00:00:00 +0000 https://ashvardanian.com/talks/ <p>Most materials are in English unless <del>literally</del> flagged otherwise. The absolute majority is on the subjects of Systems Design, Computer Science, and Artificial Intelligence. The 🗣️ talking head links aren&rsquo;t technical, and in the ones with a 👯‍♂️ - I am just a wingman supporting another speaker.</p> <h2 id="2025">2025</h2> <div style="position: relative; padding-bottom: 56.25%; height: 0; overflow: hidden;"> <iframe allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share" allowfullscreen="allowfullscreen" loading="eager" referrerpolicy="strict-origin-when-cross-origin" src="https://www.youtube-nocookie.com/embed/bDRo7Cf7x1o?autoplay=0&controls=1&end=0&loop=0&mute=0&start=0" style="position: absolute; top: 0; left: 0; width: 100%; height: 100%; border:0;" title="YouTube video" ></iframe> </div> <ul> <li>PyTorch &amp; Lightning AI Meetup: Matrix Multiplication Assembly Instructions. London, UK. <a href="https://www.meetup.com/london-pytorch-meetup/events/305522836">Event</a>. <a href="https://youtu.be/bDRo7Cf7x1o">YouTube</a>.</li> <li>OpenUK 2025: Linux Kernel 6 - Path to 10x Faster Databases and Networking. London, UK. <a href="https://youtu.be/xzFYuTqGd6o">YouTube</a>.</li> <li>PyTorch Meetup 21 at Databricks: Scaling Vector Search Up &amp; Out with Spark. London, UK. <a href="https://www.meetup.com/london-pytorch-meetup/events/310701289">Event</a>.</li> <li>OpenUK 2025: Democratizing AI - Optimizing Matrix Multiplications on Intel, Apple, AMD, NVIDIA &amp; AWS Chips. London, UK. <a href="https://youtu.be/Qujweq6Gf5A">YouTube</a>.</li> <li>PyData London 101: USearch - the Search Engine Behind Most New RAG Pipelines. London, UK. <a href="https://www.meetup.com/pydata-london-meetup/events/311635959">Event</a>.</li> <li>C++ &amp; Rust Cross-over Meetup: Writing &ldquo;Less Slow&rdquo; C++. London, UK. <a href="https://www.meetup.com/cpplondon/events/310272299">Event</a>.</li> </ul> <h2 id="2024">2024</h2> <div style="position: relative; padding-bottom: 56.25%; height: 0; overflow: hidden;"> <iframe allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share" allowfullscreen="allowfullscreen" loading="eager" referrerpolicy="strict-origin-when-cross-origin" src="https://www.youtube-nocookie.com/embed/LPv7QybMPxU?autoplay=0&controls=1&end=0&loop=0&mute=0&start=0" style="position: absolute; top: 0; left: 0; width: 100%; height: 100%; border:0;" title="YouTube video" ></iframe> </div> <ul> <li>ScyllaDB P99 Conf: Internet-Scale Semantic, Structural, and Text Search in Real Time. <a href="https://www.p99conf.io/">Event</a>. <a href="https://youtu.be/yn87sxRsOj0">YouTube</a>.</li> <li>UC Berkeley Open Source AI Day: Open Source Search: Two Decades of Bad Design Decisions &amp; Legacy Software. <a href="https://lu.ma/71y9vyb2?tk=7HrekN">Event</a>. <a href="https://youtu.be/LPv7QybMPxU">YouTube</a>.</li> <li>Arize Observe 2024. <a href="https://arize.com/observe-2024/">Event</a>.</li> <li>Neural Search Podcast: JIT Assembly to Build Exascale AI Infrastructure. <a href="https://youtu.be/zFq4-198OpQ">YouTube</a>.</li> <li>Attention Heads Podcast: Large scale data processing for AI apps. <a href="https://youtu.be/p96nkMM7wnM">YouTube</a>.</li> <li>Zaiste Programming Podcast: Unlocking the Future of AI with Open Source. <a href="https://youtu.be/D7pCY2ySicM">YouTube</a>.</li> <li>Multimodal Visual Question Answering. <a href="https://voxel51.com/computer-vision-events/may-30-2024-ai-machine-learning-data-science-meetup/">Event</a>. <a href="https://youtu.be/q_WWtZp4vgg">YouTube</a>. 👯‍♂️</li> </ul> <h2 id="2023">2023</h2> <div style="position: relative; padding-bottom: 56.25%; height: 0; overflow: hidden;"> <iframe allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share" allowfullscreen="allowfullscreen" loading="eager" referrerpolicy="strict-origin-when-cross-origin" src="https://www.youtube-nocookie.com/embed/UMrhB3icP9w?autoplay=0&controls=1&end=0&loop=0&mute=0&start=0" style="position: absolute; top: 0; left: 0; width: 100%; height: 100%; border:0;" title="YouTube video" ></iframe> </div> <ul> <li>AI.dev Linux Foundation Conference: Retrieval Augmentation and Semantic Search at Scale. San Jose, California, US. <a href="https://aidevcass23.sched.com/event/1VRvK">Event</a>, <a href="https://youtu.be/ODTpIbJ-Vks">YouTube</a>.</li> <li>All Things Open Conference: Bird&rsquo;s Eye View of Open-Source AI Infrastructure. Raleigh, North Carolina, US. <a href="https://2023.allthingsopen.org">Event</a>. <a href="https://drive.google.com/file/d/12bSVtE0ruQ_6lQSRNZN6juS2ybYL9T3S/view?usp=sharing">Slides</a>, <a href="https://www.youtube.com/watch?v=PQKYc0zK0iU">YouTube</a>.</li> <li>UC Berkeley Sky Lab Seminar: Vector Search at Scale - Bottlenecks and Solutions. Berkeley, California, US. <a href="https://sky.cs.berkeley.edu/events/sky-seminar-ashot-vardanian-unum-vector-search-at-scale-bottlenecks-and-solutions/">Event</a>.</li> <li>Cloud Native London: Future of Open-Source AI Infrastructure. London, UK. <a href="https://www.meetup.com/cloud-native-london/events/292727770/">Event</a>.</li> <li>Highload++ Conference: Vector Search and Databases at Scale. Belgrade Serbia. <a href="https://highload.rs/2023/abstracts/9770">Event</a>, <a href="https://drive.google.com/file/d/11M51Jw9UdEmzHDTGZmn4n3bxgTcQt3sw/view?usp=sharing">Slides</a>, <a href="https://www.youtube.com/watch?v=UMrhB3icP9w&amp;t=65s">YouTube</a>.</li> <li>Python SF Group: Accelerated Datascience Libraries and Where to Find Them. Sentry HQ, San Francisco, California, US. <a href="https://www.meetup.com/sfpython/events/skwnctyfchbnb/">Event</a>, <a href="https://youtu.be/L9ELuU3GeNc">YouTube</a>.</li> <li>CppCast Podcast #359: On AI Infrastructure. <a href="https://cppcast.com/ai_infrastructure/">Event</a>. <a href="https://open.spotify.com/episode/0GBepAsjOFerG9bT20WmFw">Spotify</a>, <a href="https://podcasts.apple.com/us/podcast/ai-infrastructure/id968703120?i=1000611006222">Apple Podcasts</a>, <a href="https://podcasts.google.com/feed/aHR0cHM6Ly9jcHBjYXN0LmNvbS9mZWVkLnJzcw/episode/ZjgwNzRkZmItNTUzNC00N2QwLWFjNDctZmNmNGRmMTc1NzQ1">Google Podcasts</a>.</li> </ul> <div style="position: relative; padding-bottom: 56.25%; height: 0; overflow: hidden;"> <iframe allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share" allowfullscreen="allowfullscreen" loading="eager" referrerpolicy="strict-origin-when-cross-origin" src="https://www.youtube-nocookie.com/embed/PQKYc0zK0iU?autoplay=0&controls=1&end=0&loop=0&mute=0&start=0" style="position: absolute; top: 0; left: 0; width: 100%; height: 100%; border:0;" title="YouTube video" ></iframe> </div> <ul> <li>Amazon AWS Podcast #32: Can you make a database 6-7 times faster? <a href="https://open.spotify.com/episode/7CsVxDOq8nYEBrQxRExM56&amp;nd=1">Spotify</a>, <a href="https://www.podbean.com/media/share/pb-vqhii-13e0414">Podbean</a>, <a href="https://podcasts.apple.com/by/podcast/032-%D0%BC%D0%BE%D0%B6%D0%BD%D0%BE-%D0%BB%D0%B8-%D1%83%D1%81%D0%BA%D0%BE%D1%80%D0%B8%D1%82%D1%8C-%D0%B1%D0%B0%D0%B7%D1%83-%D0%B4%D0%B0%D0%BD%D0%BD%D1%8B%D1%85-%D0%B2-6-7-%D1%80%D0%B0%D0%B7/id1600771698?i=1000608732451">Apple Podcasts</a>, <a href="https://podcasts.google.com/feed/aHR0cHM6Ly9mZWVkLnBvZGJlYW4uY29tL2F3c25hcnVzc2tvbS9mZWVkLnhtbA/episode/YXdzbmFydXNza29tLnBvZGJlYW4uY29tL2M0YjY3NzZiLTdkN2MtMzEzYS1hYzQwLWI5ZWIyNjUwM2ExMw?sa=X&amp;ved=0CAUQkfYCahcKEwjAjJ_X6bD-AhUAAAAAHQAAAAAQAQ">Google Podcasts</a>. 🇷🇺</li> <li>Attention Heads Podcast #4: Large scale data processing for AI apps. <a href="https://youtu.be/p96nkMM7wnM">YouTube</a>.</li> <li>DigiTech Conference: On deep tech startups and the development of the industry in Armenia. Yerevan, Armenia. <a href="https://www.digitec.am/ashot-vardanian">Event</a>. <a href="https://youtube.com/watch?v=V3L-O_qv7uU">YouTube</a>. 🗣️</li> <li>EVN Disrupt Podcast: Building AI While Embracing the Unorthodox. <a href="https://evnreport.com/podcasts/evn-disrupt/ashot-vardanyan-building-ai-while-embracing-the-unorthodox/">Event</a> podcast. <a href="https://www.youtube.com/watch?v=mvpyJLW2lZI">YouTube</a>. <a href="https://open.spotify.com/episode/71wdHxO4oqfRCUsoaF1J0D">Spotify</a>. <a href="https://podcasts.apple.com/am/podcast/ashot-vardanyan-building-ai-while-embracing-the/id1252212513?i=1000596262592">Apple Podcasts</a>. 🗣️</li> <li>Voxel 51 Meetup: Scaling Similarity Search with USearch. <a href="https://voxel51.com/blog/recapping-the-ai-machine-learning-and-data-science-meetup-dec-7-2023/">Event</a>. <a href="https://youtu.be/x10DpjeegQk">YouTube</a>.</li> </ul> <h2 id="2022">2022</h2> <div style="position: relative; padding-bottom: 56.25%; height: 0; overflow: hidden;"> <iframe allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share" allowfullscreen="allowfullscreen" loading="eager" referrerpolicy="strict-origin-when-cross-origin" src="https://www.youtube-nocookie.com/embed/ybWeUf_hC7o?autoplay=0&controls=1&end=0&loop=0&mute=0&start=0" style="position: absolute; top: 0; left: 0; width: 100%; height: 100%; border:0;" title="YouTube video" ></iframe> </div> <ul> <li><strong>Designing the fastest ACID Key-Value Store</strong> @ <a href="https://highload.am/2022/abstracts/9673">Highload++</a>. <a href="https://www.youtube.com/watch?v=ybWeUf_hC7o">YouTube</a>. <a href="https://drive.google.com/file/d/16gdazzt9DTpCWPuXBAV2JSe1NW-Y7HvQ/view">Slides</a>.</li> <li><strong>3M: Prospects and Challenges with Multi-Modal Models in AI Research</strong> @ <a href="https://datafest.am">DataFest</a>. <a href="https://youtu.be/p3RMkiqd7vY">YouTube</a>. <a href="https://drive.google.com/file/d/166UgMRVM1ORJPWQ74oRc2UH-bKmPHbqI/view">Slides</a>.</li> <li><strong>Accelerated Data Science Libraries</strong> @ <a href="https://pydata.org/yerevan2022/">PyData Conference</a>. <a href="https://youtu.be/OxAKSVuW2Yk">YouTube</a>. <a href="https://drive.google.com/file/d/168_Ctx0n6Jtw7ufSlTL3skCZR--lw-C0/view">Slides</a>.</li> <li><strong>Life Altering Technologies</strong> @ <a href="https://fast.foundation/gif/2022/">Global Innovation Forum</a>. <a href="https://www.youtube.com/watch?v=EBh9_7o31bI&amp;t=24447s">YouTube</a>. 🗣️</li> <li><strong>Persistent Memory Technologies Overview</strong> @ <a href="https://amd.com">AMD</a> &amp; <a href="https://www.xilinx.com">Xilinx</a>.</li> <li><strong>Role of C++ in Machine Learning discussion</strong> @ CppRussia. <a href="https://youtu.be/gO_bVvIN7HM">YouTube</a>. 🇷🇺</li> <li>From OpenCL, Thrust &amp; CUB to raw CUDA Kernels &amp; SyCL @ CppArm Meetup #3. <a href="https://github.com/unum-cloud/ParallelReductions">GitHub</a>.</li> <li>Fast Inference for Large Language Models with Vladimir Orshulevich @ PyData Meetup #2. <a href="https://youtu.be/tKwL-Q7INnQ">YouTube</a>. 👯‍♂️</li> <li>Unsafe Math, GCC Attributes, and Nifty Tricks for Google Benchmark @ CppArm Meetup #4. <a href="https://github.com/ashvardanian/BenchmarkingTutorial">GitHub</a>.</li> <li>A Practical Approach to Error Handling by Arno Schödl @ CppRussia. <a href="https://youtu.be/zNbmFRaetTA">YouTube</a>. 👯‍♂️</li> <li>Accelerated Data-Science Tools Overview @ PYerevan Meetup #16. <a href="https://youtu.be/coTgcwnzvAg">YouTube</a>.</li> <li>Bindings 101: CPython, cGo, and Java Native Interface @ CppArm Meetup #5. <a href="github.com/unum-cloud/ustore">GitHub</a>, <a href="https://youtu.be/psmfAg1Nc3s">YouTube</a>.</li> </ul> <h2 id="2021">2021</h2> <ul> <li><strong>On High-Performance Computing</strong> @ Mars Podcast. <a href="https://youtu.be/yK4Bd-6Mxk0">YouTube</a>. 🇦🇲</li> <li>Evolution: C++11, 14, 17, 20, 23, 26? @ CppArm Meetup #2. <a href="https://youtu.be/jtttoxkjTIA">YouTube</a></li> <li>Peta-Scale software in 2021 @ Code Republic. <a href="https://youtu.be/8R-43hfnPHI">YouTube</a>. <a href="https://drive.google.com/file/d/166nCWQH1-5KIPNmN4rUzebAW7mi6_eY5/view">Slides</a> 🇦🇲</li> <li>On the Gituzh Scientific Initiative @ FM106.5. <a href="https://youtu.be/89eDghXaZjI">YouTube</a>. 🇦🇲 🗣️</li> <li>SIMD with EVE by Denis Yaroshevskiy @ CppRussia. <a href="https://youtu.be/CV0e-2a_dTI">YouTube</a>. 👯‍♂️</li> </ul> <h2 id="2020">2020</h2> <ul> <li><strong>SIMD = Performance you have already paid for</strong> @ <a href="https://2020.cppconf-piter.ru/2020/spb/talks/23g3egeumhe3p4fd66pbar/">CppRussia Conference</a>. <a href="https://github.com/ashvardanian/SubstringSearchBenchmark">GitHub</a>. <a href="https://youtu.be/6Sh9QWdzo58">YouTube</a>. <a href="https://drive.google.com/file/d/16BsyqGWjpNfqG0vAb21l0eySbChC_njJ/view">Slides</a>. 🇷🇺</li> <li><strong>SIMD. Frequency Scaling Licenses and Speculative Execution</strong> @ CppArm Meetup #1. <a href="https://github.com/ashvardanian/CppBenchSubstrSearch">GitHub</a>, <a href="https://youtu.be/ft51yJ9mDcc?t=140">YouTube</a>.</li> <li>Conversing about High-Performance Computing @ Pure Virtual Cast #4. <a href="https://youtu.be/dCdBFB4LDjw">YouTube</a>. 🇷🇺</li> <li>Artsakh Must Be Independent. <a href="https://youtu.be/sN8CsCgDlHY">YouTube</a>. 🗣️</li> </ul> <h2 id="2019">2019</h2> <div style="position: relative; padding-bottom: 56.25%; height: 0; overflow: hidden;"> <iframe allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share" allowfullscreen="allowfullscreen" loading="eager" referrerpolicy="strict-origin-when-cross-origin" src="https://www.youtube-nocookie.com/embed/AA4RI6o0h1U?autoplay=0&controls=1&end=0&loop=0&mute=0&start=0" style="position: absolute; top: 0; left: 0; width: 100%; height: 100%; border:0;" title="YouTube video" ></iframe> </div> <ul> <li><strong>Deep dive into GPGPU programming</strong> @ <a href="https://cppconf-piter.ru/en/2020/spb/talks/23g3egeumhe3p4fd66pbar/?fbclid=IwAR26hl3tEhw1os0J6oLzsVPTOAuSGkZIMzwq689tEq8NH5_V7b3MHV8f_zU">CppRussia Conference</a>. <a href="https://github.com/ashvardanian/SandboxGPUs">GitHub</a>. <a href="https://youtu.be/AA4RI6o0h1U">YouTube</a>. <a href="https://drive.google.com/file/d/16AicAl99t3ZZFnza04Wnw_Vuem0w8lc7/view">Slides</a>.</li> <li><strong>AI and Computational Graphs in C++</strong> @ <a href="https://www.meetup.com/cpp-bay-area/events/261294493/">CppBayArea</a>. <a href="https://github.com/ashvardanian/NeuralSTL">GitHub</a>.</li> <li>Efficient GPGPU Programming @ <a href="https://www.jetbrains.com">JetBrains</a> HQ. <a href="https://github.com/ashvardanian/SandboxGPUs">GitHub</a>, <a href="https://youtu.be/BUtHOftDm_Y">YouTube</a>.</li> </ul>