Roman GrebennikovinNixiesearchNixiesearch: running Lucene over S3, and why we’re building a new serverless search engine.A new search engine in 2024? Yes, but stateless — index on S3, serverless — no cluster state, and with all Lucene features out of the box.Oct 10Oct 10
Roman GrebennikovinNixiesearchHow to compute LLM embeddings 3X faster with model quantizationRunning LLM embedding models is slow on CPU and expensive on GPU. We will make it up to 3X faster with ONNX model quantization, see how…Nov 13, 20231Nov 13, 20231
Roman GrebennikovinMetarankFrom zero to semantic search embedding modelA series of articles on building an accurate Large Language Model for neural search from scratch. We’ll start with BERT and…Jun 12, 202315Jun 12, 202315
Roman GrebennikovinMetarankSolving a search cold-start problem with aggregated CTRHow do you rank a new item in search results when you have no visitor feedback? It was never displayed, so it got zero clicks. A recent…Mar 14, 2023Mar 14, 2023
Roman GrebennikovinMetarankSearch results diversification with MetarankA recent Metarank 0.6.3 release introduced a new diversity feature extractor, which measures how a particular item differs from others in…Mar 7, 2023Mar 7, 2023
Roman GrebennikovinMetarankMetarank 0.6.2Only a few days ago, we’ve launched the 0.6 release that adds recommendations, and we’re back with a few fixes that our beloved users have…Feb 7, 2023Feb 7, 2023
Roman GrebennikovinMetarankMetarank 0.5.9: performance and memory improvementsWe’re pleased to announce a new version of Metarank. This time we were focused on performance and overall costs of running the service.Nov 2, 2022Nov 2, 2022
Roman GrebennikovinMetarankWhy you should post your story to HackerNews on the weekend (and it should be in Rust).We collected a dataset of HN posting and ranking activity to understand the best time to submit your own story. It’s the weekend, but we’re…Oct 15, 20221Oct 15, 20221
Roman GrebennikovinMetarankYour ML setup is not unique: you don’t need more data scientistsWe’ve been long working on diverse set of ML projects, and we see the same decisions taken and same mistakes made again and again. But ML…Sep 16, 20221Sep 16, 20221
Roman GrebennikovinMetarankMetarank RAM usage benchmarkUPDATE 14 Oct 2022: metarank 0.5.6 has optimized memory usage, check the updated numbers here.Sep 9, 2022Sep 9, 2022
