Posted by david927 6/29/2025
Ask HN: What Are You Working On? (June 2025)
It's B2B only - can't register with a free email provider, gotta own a real domain -Therefore identities are collective - companies, not company employees --Therefore all interactions are persisted at the org level rather than assigned to individual inboxes
-It allows you not just to talk but also to work together on contracts. We built a contract parser that turns contract clauses into smart, plain language objects
We're calling it Geneva and doing a friends/family/acquaintances exploratory release as I type this.
http://genevabm.com http://x.com/genevab2b https://www.linkedin.com/company/genevabm/
My secret agenda is to explore how the "information supply chain" can be tracked across the data-processing stack all the way from the original audio through transcription, the processing pipeline, and UI. I'm using language models for multi-stage summarization and want to be able to follow the provenance of summaries all the way back to the transcripts and original audio.
Yes, you could try making one using Observable Plot (which is what I used for these): https://observablehq.com/plot/transforms/dodge
One of the slides in my presentation has the full prompt I used, in case that's useful. I ran it on chunks of the podcast transcript and then merged/deduplicated the results to get the data that's visualized here.
The premise is that when I read social spaces like Reddit or X, if the government has done anything contentious you get nothing more than strident left takes, or strident right takes on the topic. Neither of which is informative or helpful.
So I have set up a site which uses AI which is specifically guided to be neutral and non-partisan, to analyses the government actions from the source documents. It then gives a summary, expected effect, benefits and disadvantages, and ranks the action against 19 "things people care about" (e.g. defence, environment, civil liberties, religious protection, etc.)
The end result is quite compelling. For example here's the page that summarises all the actions which are extremely beneficial or disadvantageous to individual liberties: https://sivic.life/tyca/tyca_individual_liberties/
The site is up now as far as I can tell. We were doing some updates a couple of hours ago which might have been when you tried it. Please have another go.
http://youtube.com/@dreamwieber
In parallel I'm working on a bunch of apps for Vision Pro -- my most well-known at the moment being Vibescape which was featured recently by Apple: https://youtu.be/QcTiDBtCafg
To round this out, my wife and I are converting a historic farm in the Pacific Northwest to regenerative agriculture practices. So far we've restored over 20 acres of native ecosystems.
If that's interesting to you there's a channel here:
I discovered that VSCode has a very nice solution so I pulled the core VSCode libraries and injected them into a Chrome extension using the dependency injection, ipc / rpc, eventing to bridge the gap between all of these isolated JS contexts and expose a single, strongly‐typed messaging API, my IPC/RPC shim sits on top of each of the native environments and communication mechanisms.
Yesterday, Microsoft released the source code for the Copilot chat. Apparently, since the basis of my Chrome extension is the same core libraries I can drop the VSCode chat UI into the side panel without much friction. Although, I might continue to use Microsoft's FluentUI chat currently implemented in the extension.
Because Copilot chat has a lot of code that runs in node in Electron, now I'm working in porting all the agent capabilities for browser automation from the Copilot chat including the code for intent, prompt creation, tools, disambiguation, chunking, embedding, ect. I'm 4 to 6 weeks away from having feature parity of Playwright for automation from a Chrome extension side panel that can do most of the inference using huggingface transformer.js locally. Nonetheless, heuristics exposed as tools such that if the intent is playing a video, all that is required is a tool that collects all the video tags and related elements with metadata. No need to use $10 in tokens to figure out which video element to play.
Yeah, I think I'm 4 to 6 weeks away from having a Copilot chat in a browser doing agent automation.
If you want to see where I'm at today, https://github.com/adam-s/doomberg-terminal.
When I did Grub the crawler back in the day, that's what I was shooting for!
If you want a jumpstart on the Playwright stuff: https://github.com/kordless/gnosis-wraith. Runs on Google Cloud Run. The UI is still in progress but you can test it here: https://wraith.nuts.services. Uses tokens to email for login.
The extension stuff is the way to go, IMHO! You can capture any page, even automatically.
Most employee engagement software is just placation for HR. When it's common that the lowest scoring question on feedback cycles is "I believe that action will be taken based on the results of this feedback," there's something fundamentally broken with how companies handle feedback, and how the tools their given enable them to react to it.
Our end goal is to help leaders and managers identify problems with trust and communication within a team. The reality is, 90% of the time, the problem lies with the leadership itself. We're trying to provide both the tools to diagnose what the problems are, and frameworks for managers to fix them.
The big trick or the language is that it doesn't hide the pipelining you have to do to up your FMax, instead, you can manually add register stages in the places they're important, and the compiler will synchronize the other paths.
A really neat trick with this pipelining system is that submodules can respond to the amount of pipelining around them (through inferring template parameters). This way the programmer really doesn't have to think about the pipelining they do add. Examples are a FIFO's almost_full treshold, inferring how many simultaneous state there needs to be for a pipelined loop, inferring the depth of BRAM shift regs, etc.
Conjtest is a policy-as-code CLI tool which allows you to run tests against common configuration file formats using Clojure. You can write policies using Clojure functions or declarative schemas against many common configuration file formats such as YAML, JSON, HCL, and many others (full list in repo).
Under the hood, it uses Babashka and SCI (Small Clojure Interpreter) to run the policies and Conftest/Go parsers for compatibility with Conftest (https://www.conftest.dev/). It’s also possible to bring your own parser or reporting engine using Babashka scripting.
The initial big pieces are in place now, I’m preparing my end of the year to talk about Conjtest and get some feedback/issues to work on.