Show HN: Zero downtime embedding model upgrades https://ift.tt/z5qmyoH

Show HN: Zero downtime embedding model upgrades People use embedding models all the time for rag/semantic retrieval. However, when a newer, more desireable model comes out, there is an expensive (both in time and computational) cost of re-embedding every document in the database. However, I figured out an interesting way to forgo that upfront embedding cost. algo: old model/index -> retrieve top-K docs -> score those docs with the new model -> cache/materialize the new embeddings so instead of rebuilding the entire vector store upfront, the old index keeps getting retrieved from, while the new model reranks those candidates. This works surprisingly well for some model pairs, (i tested 63 source-> target migrations on h100s, on upto 1M documents). For example, on a 1M document Natural Questions dataset, native Qwen3-Embedding-8B: 0.6812 nDCG@10 Qwen3-4B -> Qwen3-8B, K=50: 0.6816 Qwen3-0.6B -> Qwen3-8B, K=50: 0.6638 MiniLM -> Qwen3-8B, K=50: 0.6486 (the hard part is determining k, I held the k constant above to give some sense of migratability). You can install it with pip pip install embedflow and the code is on github https://ift.tt/AecTbXC https://ift.tt/AecTbXC September 7, 2026 at 11:35PM

Show HN: NYC MapTap – Learn NYC neighborhoods (with subway routes when you miss) https://ift.tt/f0gbWHY

Show HN: NYC MapTap – Learn NYC neighborhoods (with subway routes when you miss) https://albertjoseph0.github.io/nyc-maptap/ September 7, 2026 at 11:36PM

Show HN: Server hardening playbook where every item is failure/fix/verify https://ift.tt/aUThb4Y

Show HN: Server hardening playbook where every item is failure/fix/verify https://ift.tt/x8ZIQDt September 7, 2026 at 01:31AM

Show HN: Golden hour API and a 3D globe of the best light https://ift.tt/zPV7T18

Show HN: Golden hour API and a 3D globe of the best light https://ift.tt/Jkln5Dc September 7, 2026 at 01:18AM

Show HN: Lantunnel – a P2P-first private mesh for reaching your LANs https://ift.tt/U6oRSQc

Show HN: Lantunnel – a P2P-first private mesh for reaching your LANs https://ift.tt/4CV3w1E September 7, 2026 at 12:47AM

Show HN: Blunderbase: A Personal Chess Database https://ift.tt/L2xqFJz

Show HN: Blunderbase: A Personal Chess Database I made a free and open source chess database. It's useful for personal game study and gathering statistics across sites. Supports import from lichess, chess.com, FICS and PGN then runs Stockfish and Maia over every game. With it you can analyze your games, take notes on positions and look up openings. The self hosted web version includes a remote-runner that allows connecting engines running on other machines (finally, the CPUs of the AI server at work come to use :D) and an MCP server with a live board to push positions and comments to. Roadmap: iOS and Android companion apps, a sync service, multi-engine analysis, and a correspondence game mode supporting multiple long running calculations. Looking forward to feedback and critique. Docker, Windows, Mac: https://blunderbase.org Demo: https://ift.tt/28F9i0g (read-only) https://ift.tt/mwV3G1x September 6, 2026 at 04:21AM

Show HN: Keen Bean – Mac meeting notes that draft the spec while you talk https://ift.tt/8dw9Ehm

Show HN: Keen Bean – Mac meeting notes that draft the spec while you talk https://keenbean.app/ September 6, 2026 at 06:53AM