Hacker News: chonghaoju

New comment by chonghaoju in "Ask HN: What is your database size"

chonghaoju — Sun, 09 Aug 2026 02:31:21 +0000

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.

New comment by chonghaoju in "Show HN: RelativeDB – OSS query engine for relational foundation models"

chonghaoju — Mon, 27 Jul 2026 03:46:22 +0000

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.

New comment by chonghaoju in "Show HN: Axtary – Content Authorization for AI Agents"

chonghaoju — Sun, 26 Jul 2026 03:58:32 +0000

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.

New comment by chonghaoju in "Show HN: Openleetcode – local LeetCode runner where tests live in the repo"

chonghaoju — Sat, 25 Jul 2026 04:06:04 +0000

Keeping test cases in the repo is the part that actually matters. I've lost hours to LeetCode silently changing edge cases between sessions.

New comment by chonghaoju in "Show HN: Turo – An Aggressive Token-Saving Proxy for CLI AI Agents"

chonghaoju — Fri, 24 Jul 2026 04:14:37 +0000

curious how it handles tool call schemas, those can be surprisingly token-heavy, sometimes dwarfing the actual prompt content.

New comment by chonghaoju in "One Wikipedia page costs your AI agent 68,000 tokens"

chonghaoju — Sat, 11 Jul 2026 02:51:31 +0000

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.

New comment by chonghaoju in "Show HN: FactIQ – a realtime econ+finance database for AI agents"

chonghaoju — Wed, 08 Jul 2026 04:21:08 +0000

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?

New comment by chonghaoju in "Chinese proverb: If you want to be happy for an hour, take a nap"

chonghaoju — Tue, 07 Jul 2026 06:27:02 +0000

If you want to live longer, take a nap but less than 20 min.

New comment by chonghaoju in "Show HN: LLM Thought Visualization"

chonghaoju — Tue, 07 Jul 2026 06:25:15 +0000

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?

New comment by chonghaoju in "Show HN: FCS – an open protocol for splitting contributions in pre-company teams"

chonghaoju — Tue, 07 Jul 2026 05:22:38 +0000

Interesting idea. Btw, I see 灵活就业规范 in en version.

New comment by chonghaoju in "12.4M US business registrations are free on state open-data portals"

chonghaoju — Tue, 07 Jul 2026 04:13:24 +0000

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)

12.4M US business registrations are free on state open-data portals

chonghaoju — Tue, 07 Jul 2026 04:12:56 +0000

Article URL: https://gist.github.com/BhackerJ/22a35fead8764782189c7a20c5ded0d7

Comments URL: https://news.ycombinator.com/item?id=48813596

Points: 4

# Comments: 2

New comment by chonghaoju in "GLM 5.2 beats Claude in our benchmarks"

chonghaoju — Mon, 29 Jun 2026 06:48:50 +0000

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.

New comment by chonghaoju in "Show HN: The CLI for browser agents"

chonghaoju — Mon, 29 Jun 2026 03:02:00 +0000

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

New comment by chonghaoju in "Show HN: Noise Lang, JIT stochastic programming language in Rust"

chonghaoju — Sat, 27 Jun 2026 07:47:06 +0000

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.

New comment by chonghaoju in "AI agents are sensitive to nudges"

chonghaoju — Fri, 26 Jun 2026 04:37:53 +0000

Prompt wording that implies the agent "should already know" something tends to suppress clarifying questions and leads to silent wrong assumptions