Hacker News: jovezhong

New comment by jovezhong in "MLX-based LLM inference engine for macOS with native Swift implementation"

jovezhong — Fri, 06 Jun 2025 06:14:34 +0000

Another cool option to run local models on macOS. LM Studio might have a better user interface and download experience, but Swama’s pure CLI experience is still awesome!

MLX-based LLM inference engine for macOS with native Swift implementation

jovezhong — Fri, 06 Jun 2025 06:14:34 +0000

Article URL: https://github.com/Trans-N-ai/swama

Comments URL: https://news.ycombinator.com/item?id=44198236

Points: 2

# Comments: 1

New comment by jovezhong in "[dead]"

jovezhong — Fri, 25 Apr 2025 17:46:37 +0000

Blog: https://hackaday.com/2025/04/23/abusing-duckdb-wasm-to-creat... Repo: https://github.com/patricktrainer/duckdb-doom

a great way to abuse DuckDB

New comment by jovezhong in "GitHub MCP Server"

jovezhong — Sat, 05 Apr 2025 22:10:07 +0000

I am a bit surprised the recommended/only guide to use this mcp server is via docker run, although running via node should be possible too https://github.com/github/github-mcp-server/issues/111

New comment by jovezhong in "The Llama 4 herd"

jovezhong — Sat, 05 Apr 2025 22:06:46 +0000

why only llama3.x models are listed on ollama? llama4 no longer wants to support ollama, to better track the adoption?

New comment by jovezhong in "A Year of Rust in ClickHouse"

jovezhong — Wed, 02 Apr 2025 16:11:21 +0000

"Sadly, ClickHouse is written in C++, not in Rust.".. nice opening

New comment by jovezhong in "WarpStream marketing website with a terminal interface"

jovezhong — Tue, 01 Apr 2025 17:45:21 +0000

Certainly fun to use. Maybe you will check the docs/demo more seriously, after you type the command in the UI..

WarpStream marketing website with a terminal interface

jovezhong — Tue, 01 Apr 2025 17:45:21 +0000

Article URL: https://www.warpstream.com/etc/terminal

Comments URL: https://news.ycombinator.com/item?id=43549587

Points: 1

# Comments: 1

New comment by jovezhong in "FireScale: Benchmark from Firebolt, simulates real-world production workloads"

jovezhong — Fri, 28 Mar 2025 05:21:14 +0000

The results show Firebolt delivering: 8x better price-performance than Snowflake, 18x better than Redshift, and 90x better than BigQuery. Near-linear concurrency scaling, achieving low latency of 120 milliseconds with a throughput of 2500 QPS

FireScale: Benchmark from Firebolt, simulates real-world production workloads

jovezhong — Fri, 28 Mar 2025 05:21:14 +0000

Article URL: https://github.com/firebolt-db/benchmarks

Comments URL: https://news.ycombinator.com/item?id=43501822

Points: 1

# Comments: 1

New comment by jovezhong in "TDgpt: AI agent for advanced time-series analytics"

jovezhong — Thu, 27 Mar 2025 01:53:19 +0000

SELECT _flow, _fhigh, _frowts, FORECAST(i32, "algo=lstm,model=lstm_foo,alpha=95,period=10,wncheck=0") FROM foo;

TDgpt: AI agent for advanced time-series analytics

jovezhong — Thu, 27 Mar 2025 01:53:19 +0000

Article URL: https://tdengine.com/tdgpt/

Comments URL: https://news.ycombinator.com/item?id=43489646

Points: 2

# Comments: 1

New comment by jovezhong in "Show HN: macOS app to help you blink more"

jovezhong — Sat, 22 Mar 2025 16:01:24 +0000

I was wondering how to figure out "when you stare at it for too long", then saw "Uses Mac's built-in front facing camera".. Err, I am not sure whether I want to keep camera on for this..

New comment by jovezhong in "Ask HN: Are You Polite to AI?"

jovezhong — Fri, 21 Mar 2025 16:28:05 +0000

Typing "thanks" or "great" to AI make me feel better

New comment by jovezhong in "Show HN: MCP-Kafka – Natural Language Interface for Kafka Commands"

jovezhong — Thu, 20 Mar 2025 17:34:10 +0000

Claude can help you write a lot of okay Python code. Yes, Go is great, but in this MCP case, because of uvx and pypi, as well as kafka python lib, choosing Python is not bad at all

Show HN: Open-Source C++ Apache Iceberg Client with Write Support

jovezhong — Thu, 20 Mar 2025 17:31:29 +0000

Existing OSS C++ projects like ClickHouse and DuckDB support reading from Iceberg tables. Writing requires Spark, PyIceberg, or managed services.

In this PR https://github.com/timeplus-io/proton/pull/928, we are open-sourcing a C++ implementation of Iceberg integration. It's an MVP, focusing on REST catalog and S3 read/write(S3 table support coming soon). You can use Timeplus to continuously read data from MSK and stream writes to S3 in the Iceberg format. No JVM. No Python. Just a low-overhead, high-throughput C++ engine. Docker/K8s are optional. Demo video: https://www.youtube.com/watch?v=2m6ehwmzOnc

Help us improve the code to add more integrations and features. Happy to contribute this to the Iceberg community. Or just roast the code. We’ll buy the virtual coffee.

Comments URL: https://news.ycombinator.com/item?id=43426217

Points: 8

# Comments: 3

New comment by jovezhong in "Show HN: MCP-Kafka – Natural Language Interface for Kafka Commands"

jovezhong — Thu, 20 Mar 2025 05:57:39 +0000

Nice. I built something similar but with Python https://github.com/jovezhong/mcp-timeplus

with pipy and uvx, it will be a bit easier for others to install the mcp server. Just to add a JSON, without need to download binary manually.

Show HN: Explain stream processing watermark with a react comp in docs, by AI

jovezhong — Sun, 16 Mar 2025 18:51:07 +0000

We all know AI can generate code and content. (maybe just me), I just realized AI can also generate code *for* documentation, to create easy-to-understand and interactive visualization/tool to better explain complex mechanism.

For example, in the stream processing space, watermark is not quite easy to understand. In the past I have a few blogs/talks about this but with Claude or any other capable model, you can ask AI to create an interactive demo with configurable settings, random data and frame-by-frame explanation.

Live Demo: https://docs.timeplus.com/understanding-watermark#try-it-out Source Code (80% AI,20% me): https://github.com/timeplus-io/docs/blob/main/src/components...

docusaurus and other docs framework supports React components for a long time, but only experienced frontend developers can build such interactive demo. Now with AI, you can explain the concepts to the AI and let it create the initial interactive tool. For sure there will be many issues in the initial code. You have to keep correcting them, giving then suggestions and refine or even update the code by yourself. I spent ~4 hours working with AI to refine this tool. If I wrote from scratch I probably need 20+ hours

Comments URL: https://news.ycombinator.com/item?id=43381211

Points: 1

# Comments: 0

New comment by jovezhong in "Show HN: MCPGod: Fine-grained control over MCP clients, servers, and tools"

jovezhong — Sat, 15 Mar 2025 05:32:32 +0000

Running a mcp tool is expected to be lightweight. Starting a docker container is not impossible but will make this process a bit heavy. Maybe in the future the MCP client can provide python/nodejs runtime and also have extra flag to allow the users to confirm the requested permissions for certain mcp tools. Today running MCP servers with whatever executable available locally is too risky

New comment by jovezhong in "Show HN: MCPGod: Fine-grained control over MCP clients, servers, and tools"

jovezhong — Fri, 14 Mar 2025 23:23:47 +0000

most of the MCP tools use Python env (uvx) or Node or even Java to run ANY CODE on your machine, so even the python virtual env is a sandbox but it's to isolate the dependencies not the file/network access. If you are unlucky, you can still install a malware mcp server to clean up your disk or send your photos to somewhere. MCP servers are just local scripts. There are some permission control from deno but this is not the only runtime engine for MCP server. It'll be cool to have something like Chrome extension permission or iOS/Android permission ask, but I highly doubt this will be available since on your local server, there are just too many ways to run scripts.