chonghaoju

7 ms·

1.
by chonghaoju 1mo ago
At ~35TB Postgres we ended up separating pg_wal onto its own volume first — write amplification there was quietly eating 20–30% of our NVMe throughput before we caught it.
2.
by chonghaoju 2mo ago
Curious how it handles foreign key traversal at inference time, in my experience that's where tabular models tend to fall apart when the join graph isn't seen during pre training.
3.
by chonghaoju 2mo ago
How do you deal with approval fatigue? In our experience users start rubber-stamping everything within a week, and then the whole layer is theater.
4.
by chonghaoju 2mo ago
Keeping test cases in the repo is the part that actually matters. I've lost hours to LeetCode silently changing edge cases between sessions.
5.
by chonghaoju 2mo ago
curious how it handles tool call schemas, those can be surprisingly token-heavy, sometimes dwarfing the actual prompt content.
6.
by chonghaoju 2mo ago
stripping to markdown with Jina Reader or Trafilatura before passing to the agent cuts that 68k down to ~3-5k for most Wikipedia pages, and handles the JS-rendered case too.
7.
by chonghaoju 2mo ago
Works well alongside my Robinhood MCP server! Do you store point in time vintages or only latest values? And what's pricing after the free tier?
8.
by chonghaoju 2mo ago
If you want to live longer, take a nap but less than 20 min.
9.
by chonghaoju 2mo ago
The moment that grabs me is that the verdict is internally settled several tokens before "not correct" gets verbalized. What's the per-token cost of reading the lens at nine depths?
10.
by chonghaoju 2mo ago
Interesting idea. Btw, I see 灵活就业规范 in en version.
11.
by chonghaoju 2mo ago
Author here. The scraper is ~300 lines of Python stdlib. Happy to answer questions about the rate limits or the states I tried and couldn't get (Delaware was the most surprising — zero bulk access at any price)
12.
12.4M US business registrations are free on state open-data portals(gist.github.com)
4 points by chonghaoju 2mo ago|2 comments
13.
by chonghaoju 3mo ago
Every agent run writes an audit record. Not for compliance theater — because when something breaks at 2am, you need to know exactly what happened and why.
14.
by chonghaoju 3mo ago
curious how you handle sites that fingerprint headless browsers, most of the hard work in browser automation ends up being anti-bot evasion rather than the actual task execution
15.
by chonghaoju 3mo ago
Representing variables as distributions rather than point values is underrated for telecom noise modeling — most engineers just slap a variance on top after the fact and call it done.
16.
by chonghaoju 3mo ago
Prompt wording that implies the agent "should already know" something tends to suppress clarifying questions and leads to silent wrong assumptions