Show HN: I scraped 3B Goodreads reviews to train a better recommendation model https://ift.tt/hiJ7SKz

Show HN: I scraped 3B Goodreads reviews to train a better recommendation model Hi everyone, For the past couple months I've been working on a website with two main features: - https://book.sv - put in a list of books and get recommendations on what to read next from a model trained on over a billion reviews - https://ift.tt/Hp3JtuE - put in a list of books and find the users on Goodreads who have read them all (if you don't want to be included in these results, you can opt-out here: https://ift.tt/IqGHoad ) Technical info available here: https://ift.tt/XijL93T Note 1: If you only provide one or two books, the model doesn't have a lot to work with and may include a handful of somewhat unrelated popular books in the results. If you want recommendations based on just one book, click the "Similar" button next to the book after adding it to the input book list on the recommendations page. Note 2: This is uncommon, but if you get an unexpected non-English titled book in the results, it is probably not a mistake and it very likely has an English edition. The "canonical" edition of a book I use for display is whatever one is the most popular, which is usually the English version, but this is not the case for all books, especially those by famous French or Russian authors. https://book.sv November 5, 2025 at 11:20PM

Show HN: TabPFN-2.5 – SOTA foundation model for tabular data https://ift.tt/CWPirDX

Show HN: TabPFN-2.5 – SOTA foundation model for tabular data I am excited to announce the release of TabPFN-2.5, our tabular foundation model that now scales to datasets of up to 50,000 samples and 2,000 features - a 5x increase from TabPFN v2, published in the Nature journal earlier this year. TabPFN-2.5 delivers state-of-the-art predictions in one forward pass without hyperparameter tuning across classification and regression tasks. What’s new in 2.5 : TabPFN-2.5 maintains the core approach of v2 - a pretrained transformer trained on more than hundred million synthetic datasets to perform in-context learning and output a predictive distribution for the test data. It natively supports missing values, cateogrical features, text and numerical features is robust to outliers and uninformative features. The major improvements: - 5x scale increase: Now handles 50,000 samples × 2,000 features (up from 10,000 × 500 in v2) - SOTA performance: TabPFN-2.5 outperforms tuned tree-based methods and matches the performance of a complex ensemble (AutoGluon 1.4), that itself includes TabPFN v2, tuned for 4 hours. Tuning the model improves performance, outperforming AutoGluon 1.4 for regression tasks. - Rebuilt API: New REST interface along with Python SDK with dedicated fit & predict endpoints, making deployment and integration more developer-friendly - A distillation engine that converts TabPFN-2.5 into a compact MLP or tree ensemble while preserving accuracy and offer low latency inference. There are still some limitations. The model is designed for datasets up to 50K samples. It can handle larger datasets but that hasn’t been our focus with TabPFN-2.5. The distillation engine is not yet available through the API but only through licenses (though we do show the performance in the model report). We’re actively working on removing these limitations and intend to release newer models focused on context reasoning, causal inference, graph networks, larger data and time-series. TabPFN-2.5 is available via API and a package on Hugging Face. Would love for you to try it and give us your feedback! Model report: https://ift.tt/YPvk14E... Package: https://ift.tt/yUEo5dF Client: https://ift.tt/vXxmtyr Docs: https://ift.tt/P5ebKSk https://ift.tt/G3RJaYZ November 6, 2025 at 11:56PM

Show HN: JermCAD – A YAML-powered, vibe-coded, browser-based CAD software https://ift.tt/5hWTu8z

Show HN: JermCAD – A YAML-powered, vibe-coded, browser-based CAD software I had a hard time figuring out CAD software like Fusion, OnShape, etc., and decided to go about making my own CAD modeling software that I can "program" my models similar to how I think about them in my head. I used Cursor to write like 95+% of this, giving it my YAML examples and making it implement the actual code to make those work. Currently 100% self-hosted, and it is just a static HTML/CSS/JS, so it might just work without running npm at all. Very few features working currently, basically just modeling a few primitive solids, and boolean operations. https://ift.tt/9c8U30k November 5, 2025 at 08:31PM

Wrapped and Ready for Your Holiday Fun: The Merry Days of Muni

Wrapped and Ready for Your Holiday Fun: The Merry Days of Muni
By Glennis Markison

Mittens and presents and San Francisco landmarks all wrapped around your 49 Van Ness / Mission? That’s right – the Merry Days of Muni have just begun! This festive bus wrap marks the start of a full campaign of merry moments for Muni riders. It’s our way of bringing local communities a little extra joy this holiday season. Through the end of the year, we’re finding fun ways to connect riders as they take Muni to their favorite people, places and traditions across the city. Meet the SFMTA designer behind our first holiday wrap – and see where the festive ride can take you. Muni's first holiday...

Published November 05, 2025 at 05:30AM
https://ift.tt/fSPlBFz

Show HN: ReadMyMRI DICOM native preprocessor with multi model consensus/ML pipes https://ift.tt/H4txQBC

Show HN: ReadMyMRI DICOM native preprocessor with multi model consensus/ML pipes I'm building ReadMyMRI to solve a problem I kept running into: getting medical imaging data (DICOM files) ready for machine learning without violating HIPAA or losing critical context. What it does: ReadMyMRI is a preprocessing pipeline that takes raw DICOM medical images (MRIs, CTs, etc.) and: Strips all Protected Health Information (PHI) automatically while preserving DICOM metadata integrity Compresses images to manageable sizes without destroying diagnostic quality Links deidentified scans to user-provided clinical context (symptoms, demographics, outcomes) Uses multi-model AI consensus analysis for both consumer facing 2nd opinions and clinical decision making support at bedside Outputs everything into a single dataframe ready for ML training using Daft (Eventual's distributed dataframe library) Technical approach: Built on pydicom for DICOM manipulation Uses Pillow/OpenCV for quality-preserving compression Daft integration for distributed processing of large medical imaging datasets Frontier models for multi model analysis (still debating this) What I'm looking for: Feedback from anyone working with medical imaging ML Edge cases I haven't thought about Whether the Daft integration actually makes sense for your use case or if plain pandas would be better HIPAA/privacy concerns I am not thinking about Happy to answer questions about the architecture, HIPAA considerations, or why medical imaging data is such a pain to work with. https://ift.tt/mKvQlWo November 5, 2025 at 04:17AM

Show HN: Barcable – We Built Agents That Automatically Load Test Your Back End https://ift.tt/suyI3FV

Show HN: Barcable – We Built Agents That Automatically Load Test Your Back End Hey HN, we’re Iyan and Datta, founders of Barcable. Barcable connects to your backend (HTTP, gRPC, GraphQL) and uses autonomous agents to generate and run load tests directly inside your CI/CD. No configs, no scripts. It scans your repo, understands your API routes, and builds real test scenarios that hit your endpoints with realistic payloads. Docs: https://ift.tt/aJ0N3zW We built this out of frustration. Every team we’ve worked with ran into the same issue: reliability testing never kept up with development speed. Pipelines deploy faster than anyone can validate performance. Most “load tests” are brittle JMeter relics or one-off scripts that rot after the first refactor. Barcable is our attempt to automate that. It: - Parses your OpenAPI spec or code to discover endpoints automatically - Generates realistic load tests from PR diffs (no manual scripting) - Spins up isolated Cloud Run jobs to execute at scale - Reports latency, throughput, and error breakdowns directly in your dashboard - Hooks into your CI so tests run autonomously before deploys Each agent handles a part of the process—discovery, generation, execution, analysis—so testing evolves with your codebase rather than fighting against it. Right now it works best with Dockerized repos. You can onboard from GitHub, explore endpoints, generate tests, run them, and see metrics in a unified dashboard. It’s still a work in progress. We’ll create accounts manually and share credentials with anyone interested in trying it out. We’re keeping access limited for now because of Cloud Run costs. We’re not trying to replace performance engineers, just make it easier for teams to catch regressions and incidents before production without the setup tax. Would love feedback from anyone who’s been burned by flaky load testing pipelines or has solved reliability differently. We’re especially curious about gRPC edge cases and complex auth setups. HN has always been a huge source of inspiration for us, and we’d love to hear how you’d test it, break it, or make it better. — Iyan & Datta https://ift.tt/ZkircDx https://ift.tt/36IOlEc November 5, 2025 at 04:55AM

Show HN: Agor → Figma for AI Coding (Open Source) https://ift.tt/I9AbxRW

Show HN: Agor → Figma for AI Coding (Open Source) https://agor.live November 4, 2025 at 07:29PM