Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Ask HN: What Are You Working On? (March 2026)

news.ycombinator.com

Post reply on HN
Print view View original

491–500 of 1001 posts

Re: Ask HN: What Are You Working On? (March 2026)

#491

Post by merlindru »

Tool that lets you build shortcuts and custom functionality for most macOS apps. You write JS snippets like this:

    const app = new App("com.apple.finder")
and then query for elements:

    const window = app.$({role: "window"})
    const someButton = window.$(/* another query */)
and then do stuff with it:

    someButton.press()
and you can bind everything to very specific shortcuts like "press and hold cmd, then scroll mouse wheel up"

Targeted towards music producers and AI (there's one collection of snippets that starts an MCP server and exposes some basic functionality) in the beginning.

Re: Ask HN: What Are You Working On? (March 2026)

#492

Post by marusalu »

I'm building a microreading service that let's me get long books read with small chunks of time that I have - https://lauselt.ee Currently I've added some public domain Estonian books in there and tbh I do get a lot more reading done during the day. Basically you can use your 1-5 min breaks (waiting for a bus, during the commercials, waiting for food etc) to open the book quickly where you left off and read by scrolling small chunks of texts at a time. Duolingo style streak to create the habit of reading every day. Also the ability to upload your own book and it will automatically be split into these small chunks.

Re: Ask HN: What Are You Working On? (March 2026)

#493

Post by marginalia_nu »

Cooking up a NSFW filter for marginalia search.

Pipeline so far has gone like this:

* Use the search engine's API to query a bunch of depravity

* Use qwen3.5 to label the search results and generate training data

* Try to use fasttext to create a fast model

* Get good results in theory but awful results in practice because it picks up weird features

* Yolo implement a small neural net using hand selected input features instead

* Train using fasttext training data

* Do a pretty good job

* for (;;) Apply the model to real a world link database and relabel positive findings with qwen to provide more training data

Currently this is where I'm at

  Accuracy:   90.90%
  True  Positive: 1021
  False Positive: 154
  True  Negative: 2816
  False Negative: 230
  Precision:  0.8689
  Recall:     0.8161
  F1:         0.8417
There's a lot of vague middle ground and many of the false positives are arguably just mislabeled.

Re: Ask HN: What Are You Working On? (March 2026)

#495

Post by franze »

Google AI overview dissector

AIOs are a black hole - we dont know when they appear and whats in it. so i creates a tool thats starts with GSC data and enriches it via AIO data

works good and the major finding by now

the best AIOs you can get are ..... none.

doesn't matter if you are in it or not - as soon as they show up the CTR to tour web-property goes down massively ~60% to 70%

the CTR on the AIOs are ~0%

Re: Ask HN: What Are You Working On? (March 2026)

#496

Post by paragarora »

Building HEBBS — a memory engine for AI agents, written in Rust.

The problem: every agent framework bolts together a vector DB for recall, a KV store for state, maybe a graph DB for relationships, and then hopes the duct tape holds. You get one retrieval path (similarity search), no decay, no consolidation, and the agent forgets everything the moment context gets trimmed.

HEBBS replaces that stack with a single embedded binary (RocksDB underneath, ONNX for local embeddings). Nine operations in three groups: write (remember, revise, forget), read (recall, prime, subscribe), and consolidate (reflect, insights, policy). The interesting part is four recall strategies — similarity, temporal, causal, and analogical — instead of just "nearest vector."

Some technical decisions I'm happy with:

- No network calls on the hot path. Embeddings run locally via ONNX; LLM calls only happen in the background reflect pipeline.

- recall at 2ms p50 / 8ms p99 at 10M memories on a 2 vCPU instance.

- Append-only event model for memories — sync is conflict-free, and forget is itself a logged event (useful for GDPR).

- Lineage tracking: insights link back to source memories, revisions track predecessors.

SDKs for Python, TypeScript, and Rust. CLI with a REPL. gRPC + REST.

There's a reference demo — an AI sales agent that uses HEBBS for multi-session memory, objection handling recall, and background consolidation of conversation patterns.

Still early. The part I'm wrestling with now is tuning the reflect pipeline — figuring out when and how aggressively to consolidate episodic memories into semantic insights without losing useful detail. Curious if anyone working on agent memory has opinions on that tradeoff, or if you've found other approaches that work.

https://github.com/hebbs-ai/hebbs

Re: Ask HN: What Are You Working On? (March 2026)

#497

Post by esperent »

BrunoBernardino wrote: post #485

My wife and I continue to work on Uruky, a EU-based Kagi alternative [1]. Since last month we got deals with a couple more search providers but we’re still waiting for EUSP/STAAN to provide us with an API key (we have progressed through a few more forms and signatures and legal stuff, though). We’ve continued to get some paid customers and have exited beta last week, given everyone seemed to be quite satisfied and th…

Very cool, wishing you the best of luck with this.

One bit of feedback from me, take it or leave it, but the name doesn't feel appealing or memorable. What does it mean?

Re: Ask HN: What Are You Working On? (March 2026)

#498

Post by jonaslanglotz »

I'm working on site that let's you check when a manned space station was last directly over your house.

It's a reference to https://xkcd.com/2883/, which I've always liked and was suprised there was no tool to check when you last had astronauts over for dinner.

Looking up the location of the ISS at a specific time is easy. Looking up the closest passes of the ISS to a specific location for the last 30 years on-demand is more complicated.

Re: Ask HN: What Are You Working On? (March 2026)

#500

Post by rishikeshs »

I'm building a small tool called FormBeep[1] that sends a notification to your phone when someone submits a form on your website.

It started as a client problem, then something which I also experienced so decided to built it. It's just one small script and work seamlessly across platforms.

[1]https://formbeep.com

Post reply on HN