HN user
This is really cool! Really give a immersion vibe.
I've built something different, Tokyo Train Orchestra (http://tokyo-train-orchestra.netlify.app/) It uses live and scheduled tokyo train/subway timetable to produce music.
For anyone who is interested in learning Japanese, or looking for resources. I've compiled this "Awesome Japanese" repo.
I don't like claude code web due to its lack of planning mode. I found the result is often lackluster compare to claude code cli.
My current setup: Tailscale + Terminus(ipad) + home machine(code base)
Need to look into how to work on multiple features at the same time next.
I wouldn't trust Zoho. More than 10 years ago, they shadowbanned (can not be shared or publicly viewed) my documents because it criticized Chinese communist party.
I'm building a information hub that utilize AI agents to compile all the relevant information from China (gov, econ, commerce, anything and everything). I call it "information domination". Heard many times that language barrier stops people having a better understanding of China. Now the LLM translation is good enough, and AI agents can do way more in information gathering stage. It's time to put things in practice.
For now, it only has a daily newsletter fully compiled by AI agents without any human intervention. I plan to add public listed companies (semiconductor, energy provider, etc) onto the platform. Already found lots of good data points that can be used by analysts, researchers or observers.
Q: Has anyone used claude working with DaisyUI? Don't like claude's default ui choices.
AI Anime Recommendation Engine
Do you have a benchmark for comparing different chunking methods? Your existing benchmark is to compare different libraries.
my project deals w/ non-english text, bm25 performance is middeling. Language specific sparse model helps.
have you thought about SPALDE models? ex: https://arxiv.org/abs/2109.10086
Recently tested a (non-english) pdf ocr with Gemini 2.5 Pro. First, directly ask it to extract text from pdf. Result: random text blob, not useable.
Second, I converted pdf into pages of jpg. Gemini performed exceptional. Near perfect text extraction with intact format in markdown.
Maybe there's internal difference when processing pdf vs jpg inside the model.
What method/model are you using for sparse search?
Will let you know once android version is public
Japan Top 100 (restaurants) app
https://apps.apple.com/us/app/japan-top-100/id6741251616
Only public on iOS store for now, due to Google requires larger # of testers to authenticate the app.
This app idea came from my last Japan trip. I had a hard time finding good local restaurants (not tourist traps). So I decided to build this app with Japanese-oriented lists, due to lists' lacked of maps, etc. The app actually helped me discover some great spots, since it lists places all over Japan. Please give it a try and let me know if you have any feature requests or ideas.
China's land area is approximately 25 times of Japan, def has a bigger network. But the top speed is comparable in current generations[1] except the Maglev.
1. https://www.railway-technology.com/features/the-10-fastest-h...
Can you please elaborate on the Chinese advantage?
Claude api lacks structured output, without uniformity in output, it's not useful as agent. I've had agents system broke down suddenly due to degradation in output, which leads to the previous suggested json output hacks (from official cookbook) stopped working.
Tried Agent-E, like the DOM distillation to reduce token method. I've started using AgentQL for a new project, which works well combining playwright.
``` GET https://localhost:3001/socket.io/?EIO=4&transport=polling&t=... net::ERR_CONNECTION_REFUSED ```
Looks like you should set a production ENV?
Lots of different projects. Recently finished: beatcode: https://github.com/yudataguy/beatcode
this is built w/ the help of claude 3.5 sonnet (new) and cursor. The idea came from want to space repetition memorization for leetcode.
Langchain and other frameworks are too bloated, it's good for demo, but highly recommend to build your own pipeline in production, it's not really that complicated, and you can have much better control over implementation. Plus you don't need 99% packages that comes with Langchain, reduce security vulnerabilities.
I've written a series of RAG notebooks on how to implement RAG in python directly, with minimal packages. I know it's not in Rust or C++, but it can give you some ideas on how to do things directly.
Chromecast is actually super useful for my immigrants parents. Since some foreign languages' input method on remote is horrendous, Chromecast helps them selecting youtube video easily.
In my previous position, management employed the concept of a 'hackathon competition' as a sales pitch to the judges. One team, having received advance instruction, constructed their work in progress as a hackathon demo in advance and ultimately "won". It pissed so many people once it was known.
listened to the audiobook, highly recommend.
Would love to talk about the back-office processes. Let's connect via sam@instance.co.jp
Rakuten (YuzuAI) team member which is mentioned in the article, we're in the process improving the benchmark framework to be more balanced and cost efficient.
https://github.com/microsoft/table-transformer
This looks interesting as well. Haven't tested yet.