Hacker News

Article URL: https://groupmeetingscheduler.com/

Comments URL: https://news.ycombinator.com/item?id=46975252

Points: 1

# Comments: 1

Hey HN,

I've been building a small in-memory vector search library as a way to explore ANN systems from first principles. I was inspired by Spotify's annoy, and Meta's FAISS.

Currently, it's a CPU-first C++ library with Python bindings that supports Flat/IVF indexes and Cosine/L2 distance metrics. There's a Colab notebook linked in the README if you want to try it quickly without installing anything.

I went from a naive brute-force (with millisecond level latency) to under half a millisecond with IVF, when benchmarked against SIFT1M subset. After that I was able to increase the throughput by ~2.4x, by making the search multi-threaded on my 4 core CPU (U series). Using scalar quantization, I reduced the memory usage by ~73% with negligible loss in accuracy.

I have documented the changes in performance and overall architecture in my repository.

Currently, my focus is on tightening the memory alignment and tuning cache locality, and making top-k selection faster, and in the long term my aim is to implement IVF-PQ and HNSW.

I'd appreciate any feedback on how I may move forward, and think through the process.

I used AI tools for implementing serialization and bindings and for early architecture brainstorming; details/prompts for which are documented in the README's 'Disclosure' section.

Comments URL: https://news.ycombinator.com/item?id=46975244

Points: 1

# Comments: 0

Article URL: https://restofworld.org/2026/india-ai-summit-third-way-global-south-big-tech/

Comments URL: https://news.ycombinator.com/item?id=46975239

Points: 1

# Comments: 1

Article URL: https://www.thefp.com/p/watch-the-jeffrey-epstein-tapes-part

Comments URL: https://news.ycombinator.com/item?id=46975214

Points: 2

# Comments: 0

Article URL: https://www.leemartin.com/netlify-capsules/

Comments URL: https://news.ycombinator.com/item?id=46975206

Points: 2

# Comments: 1

Article URL: https://simonwillison.net/2026/Feb/6/pydantic-monty/

Comments URL: https://news.ycombinator.com/item?id=46975196

Points: 1

# Comments: 0

Article URL: https://olix.com/blog/compute-manifesto

Comments URL: https://news.ycombinator.com/item?id=46975190

Points: 1

# Comments: 0

Article URL: https://onlybots.cam

Comments URL: https://news.ycombinator.com/item?id=46975187

Points: 1

# Comments: 0

Article URL: https://appstorelocalization.com

Comments URL: https://news.ycombinator.com/item?id=46975186

Points: 1

# Comments: 1

Article URL: https://github.com/stoatchat

Comments URL: https://news.ycombinator.com/item?id=46975183

Points: 1

# Comments: 0

Article URL: https://csmeyer.substack.com/p/originators-of-information

Comments URL: https://news.ycombinator.com/item?id=46975176

Points: 1

# Comments: 0

Article URL: https://techcrunch.com/2026/02/10/nearly-half-of-xais-founding-team-has-now-left-the-company/

Comments URL: https://news.ycombinator.com/item?id=46975169

Points: 1

# Comments: 0

Article URL: https://bart.degoe.de/building-a-semantic-search-engine-in-250-lines-of-python/

Comments URL: https://news.ycombinator.com/item?id=46975153

Points: 4

# Comments: 0

Article URL: https://babin.posthaven.com/first-to-be-second-why-enduring-ai-companies-will-be-started-after-this-wave

Comments URL: https://news.ycombinator.com/item?id=46975150

Points: 1

# Comments: 0

Article URL: https://www.heise.de/en/blog/Software-as-a-tool-not-an-end-in-itself-A-plea-for-more-domain-expertise-11168756.html

Comments URL: https://news.ycombinator.com/item?id=46975149

Points: 1

# Comments: 0

Article URL: https://www.kolega.dev/blog/why-we-found-225-security-flaws-in-45-open-source-projects-that-sast-missed/

Comments URL: https://news.ycombinator.com/item?id=46975123

Points: 1

# Comments: 1

Hi Everyone,

I’m an independent researcher (and professionally, the Global Director of Research at Reink Media) looking for an endorsement for the cs.AI (Computer Science/Artificial Intelligence) category on arXiv.

The Context I didn't start by writing a paper; I started by building a system. Over the last year, I developed a production-grade Model Context Protocol (MCP) server for the forex market. It’s currently live with 45 distinct tools and renders 28 different dynamic widgets across Claude, ChatGPT, and a custom web app.

The Paper The paper is titled Protocol-Constrained Agentic Systems: A Neuro-Symbolic Architecture for Hallucination-Resistant Financial Execution.

My core argument is that in high-stakes domains like finance, we cannot rely on LLMs to be "smart enough" to avoid critical errors. Instead, I propose an architecture that uses MCP as a "hallucination firewall." This strictly decouples the probabilistic layer (the LLM parsing intent) from the deterministic layer (the tool executing the trade). It effectively treats the protocol schema as a type system for agent actions, guaranteeing by construction that invalid tool calls cannot reach the execution layer.

You can read the full paper and see the architecture in the PDF here: https://www.stevenhatzakis.com/research/protocol-constrained-agentic-systems

The Request If you are a registered endorser for cs.AI and find this work relevant, I would appreciate your support so I can submit this paper on arXiv.org.

Endorsement Code: LZRTFH Link: https://arxiv.org/auth/endorse?x=LZRTFH

Thanks for your time.

Steven Hatzakis

Comments URL: https://news.ycombinator.com/item?id=46975121

Points: 1

# Comments: 0

Article URL: https://www.heise.de/en/blog/The-Real-Bottleneck-Why-Faster-Coding-Doesn-t-Speed-Up-Projects-11172609.html

Comments URL: https://news.ycombinator.com/item?id=46975116

Points: 2

# Comments: 0

Article URL: https://mdst.app/blog/mdst_engine_run_gguf_models_in_your_browser

Comments URL: https://news.ycombinator.com/item?id=46975112

Points: 1

# Comments: 1

Article URL: https://twitter.com/krassenstein/status/2021542392172560784

Comments URL: https://news.ycombinator.com/item?id=46974597

Points: 1

# Comments: 0