Show HN for October 6, 2025

Everyone’s chasing AI hype. I built a Google Sheet and it quietly took off.

In 2020, I made it to track my own finances for income, expenses, savings, yearly summaries etc. I shared it once on Reddit, forgot about it for a year… When I checked back, it had over 130k views and I was honestly stoked!

No launch. No funding. No AI. Just a spreadsheet people actually stick with and find useful.

I finally gave it a proper home: write-it-down.com Now, more than 2,300 people use it.

It’s intentionally boring and that’s why it works.

People don’t always need AI. They just need something that actually solves their problem. This isn’t a billion-dollar startup of course, but it taught me more about building products than almost anything else.

Build something useful. Solve a real problem. Even if it’s just a simple spreadsheet.

So, what’s the most “boring” thing you’ve built that found unexpected traction?

I built this website to host a data-driven model of my coffee sample roaster.

I realized after 20 or so batches on the machine that while the controls are intuitive (heat, fan, and drum speeds), the physics can be unintuitive. I wanted to use my historical roast data to create and tune a model that I could use to do roast planning, control, and to help me build my own intuition for roasting. This website lets you interact with my roaster in a virtual, risk-free setting!

The models are custom Machine Learning modules that honor roaster physics and bean physics (this is not GPT/transformer-based). Buncha math.

The models are trained on about a dozen real roasts. The default bean model is an Ethiopian Guji bean.

My next steps are to add other roasters and the ability to practice control/reference tracking.

Hey hn, I put together a collection of audio and agent components for Next.js built on top of ShadCN.

Hope you find this useful & let me know if there’s ant feedback a (GitHub repo will be open sourced tomorrow am)

I built it because I needed a way to see how much will my early payments save in interest overtime.

Most of the websites I found are solely made for the US, which does not help my case. So I built one that has a bit more flexibility in terms of loan data.

Any feedback is appreciated.

Excited to share a project I’ve been building for months! Would love to receive honest feedback :)

My motivation: AI is clearly going to be the interface for data. But earlier attempts (text-to-SQL, etc.) fell short — they treated it like magic. The space has matured: teams now realize that AI + data needs structure, context, and rules. So I built a product to help teams deliver “chat with data” solutions fast with full control and observability (agent tracing, quality scores, etc) — am I wrong?

The product allows you to connect any LLM to any data source with centralized context (instructions, dbt, code, AGENTS.md, Tableau) and governance. Users can chat with their data to build charts, dashboards, and scheduled reports — all via an agentic, observable loop. With slack integration as well!

* Centralize context management: instructions + external sources (dbt, Tableau, code, AGENTS.md), and self-learning * Agentic workflows (ReAct loops): reasoning, tool use, reflection * Generate visuals, dashboards, scheduled reports via chat/commands * Quality, accuracy, and performance scoring (llm judges) to ensure reliability * Advanced access & governance: RBAC, SSO/OIDC, audit logs, rule enforcement * Deploy in your environment (Docker, Kubernetes, VPC) — full control over infrastructure

GitHub: github.com/bagofwords1/bagofwords Docs / architecture / quickstart: docs.bagofwords.com

I wrote a free, proof-complete monograph on stochastic gradient methods in infinite-dimensional Hilbert spaces. It starts from the functional-analysis foundations (Riesz, Radon–Nikodym), gives five equivalent definitions of stochastic gradients, proves existence/uniqueness of the dynamics, connects the continuum limit to gradient-flow PDEs, and derives explicit convergence rates under multiple assumption sets (convex, strongly convex, PL/KL, heavy-tailed noise, proximal). It also covers special cases (Gaussian/RKHS), extends to Hilbert manifolds/Banach spaces, analyzes discretizations (stability/consistency), and includes four applications (QM ground states, elasticity, optimal control, inverse problems) with pseudocode and explicit constants. Open problems at the end invite feedback. PDF is free to redistribute if cited.

Happy to answer questions on assumptions that are actually needed in infinite-D, how the spectral picture influences rates, and what breaks outside Hilbert structure. If you want a quick start, skim the convergence summary and the applications sections.

This is my, fully bootstrapped, take on an application hosting platform. I designed it primarily for myself, to meet the following goals:

* Utilize the best parts of the cloud, leave expensive value added services behind. * Host anywhere in the world, close to your users. * Batteries included: backups, logging, monitoring, auto-scaling, containerization, databases. * Simple and user friendly.

The journey was tough. It took me round about 3 years to get everything just right. This goes against the conventional wisdom of getting something simple out quick. With a hosting platform however, in my mind, reliability, stability and certain maturity are a must have from the start.

I definitely need to work on the presentation, it's barely covering the basics. But you have to start somewhere :-)

I'd love to hear some critique from the community. Thanks!

Hey HN,

Calendar0 is:

- A system tray app that lets you view your calendar instantly without opening a browser

- An AI assistant to manage everything in your calendar using natural language commands

- Smart contact resolution which automatically converts names to emails using Google Workspace and checks all attendees' availability before booking

- One-click meeting access for smart notifications that let you join meetings instantly

I am using Gemini 2.5 Flash for NLP. Built in Electron. Trying to keep it under ~100ms response time for most commands. It was unexpectedly challenging to parse ambiguous time references ("next Monday" vs "this Monday").

Would love to have your feedback!

Thank you!

Hi everyone,

I’ve been working on Gorbes.io: a small side project that tries to rethink how we measure success.

I’m from France, where people often have mixed feelings (hate rich people to be honest) about wealth. That got me wondering: what if we ranked people not by how much they own, but by how much wealth they’ve created for others?

Instead of ranking by personal net worth (like Forbes), Gorbes measures Distributed Value — the portion of a company’s value held by others because of someone’s work or leadership.

For this first version, I built a Top 10 list of tech builders, using a simple public-data formula:

Distributed Value = Enterprise Value × (1 − Ownership%) × Role Weight

Then I compare that to their personal net worth to show how much wealth they’ve created for others.

It’s still early and experimental — I’d love feedback on: • whether this approach makes sense • ideas for better data sources • or other ways to measure “real impact”

Thanks for reading!

Sacha

Hey HN!

I built SemanticTest while working on calendar0.app (an AI calendar assistant).

While I was building the AI assistant, I noticed a lack on good AI Evals frameworks that would help me test my agent.

SemanticTest uses GPT-4 as a judge to evaluate:

- Text responses (semantic meaning)

- Tool calls (correct tools, right order)

- Multi-turn conversations

It's composable: you build tests as JSON pipelines using custom blocks.

Would love feedback. Thank you!

Hello!

Maudit is a library (not a framework) to build static websites. What this means is that a Maudit website is a normal Rust project, pages are normal Rust structs and so on.

There's a built-in function one can use that will do pages -> HTML as expected, or one can run quite literally `SomePage.build()` to build a single page.

While there are some obvious downsides in complexity on the user side, I'm hoping that this model allows people to grow past some of the limitations that traditional SSG frameworks have, where it can be hard sometimes to customise certain aspects.

It's still super early, but feedback is extremely welcome. Hope you'll like it!

Hey HN,

I've been building Hector, a declarative AI agent platform in Go that uses the A2A protocol. The idea is pretty simple: instead of writing code to build agents, you just define everything in YAML.

Want to create an agent? Write a YAML file with the prompt, reasoning strategy, tools, and you're done. No Python, no SDKs, no complex setup. It's like infrastructure as code but for AI agents.

The cool part is that since it's built on A2A (Agent-to-Agent protocol), agents can talk to each other seamlessly. You can mix local agents with remote ones, or have agents from different systems work together. It's kind of like Docker for AI agents.

I built this because I got tired of the complexity in current agent frameworks. Most require you to write a bunch of boilerplate code just to get started. With Hector, you focus on the logic, not the plumbing.

It's still in alpha, but the core stuff works. I'd love to get feedback from anyone working on agentic systems or multi-agent coordination. What pain points do you see in current approaches?

Repo: https://github.com/kadirpekel/hector

Would appreciate any thoughts or feedback!

Hi HN!

I built a 3D arcade racing game in HTML5 using PlayCanvas.

It features arcade-style physics inspired by OutRun, with PS1/Dreamcast-style graphics. The multiplayer is asynchronous: you race against the “ghosts” (replays) of up to 7 other players, and after finishing a run you can share a link with a friend so they can race your ghost 1v1. The game currently has 20 tracks, plus a “daily challenge” mode with procedurally generated tracks.

We’ve tested it on both desktop and mobile. It supports touch and keyboard controls, though I personally prefer playing with a keyboard in landscape mode. I’d love feedback on the controls, performance, and overall fun factor. And I hope to see your names on the leaderboard!

The RL ecosystem is maturing— verifiers are standardizing how we build and share environments. However, as it grows, we need observability tooling that actually understands RL primitives. Running RL experiments without visibility into rollout quality, reward distributions, or failure modes is a waste of time.

Monitor provides live tracking, per-example inspection, and programmatic access—see what's happening during runs and debug what went wrong afterward.

The future of RL observability could look like this:

You’re working alongside your model, spawning multiple versions of your environment by tweaking components at different points, much like using git worktrees for RL experiments.

The background for this is that I'm a pretty frequent X-user like I think most people here might be. I use X on safari because I think it reduces my engagement. During the lead up to the 2024 election I was getting spammed with political posts. And also I noticed the TikTokification of the website where they were pushing more and more short-form content, which I hear is still a problem. So the extension includes three things for people: 1. Filter ads (it does this in a dumb way right for now) 2. Filter videos (this has been the killer feature for me, getting rid of all the short-form video content) 3. Custom LLM filter (I've added the ability to write a custom prompt and add an OpenAI key, and it'll run it against GPT-4 mini. You might quickly hit your request per minute, so I suggest putting some money in your OpenAI key) There's was a bit of trickiness with the way Safari extensions allow network requests vs. direct Chrome extensions. The actual repo is a little messy - I'm just putting the Xcode project which I converted from a Chrome extension and did a little bit of work to get this working in Safari. I think there's lots of interesting ways this could go if anyone wanted to encourage me to add to it. But yeah, just putting it out there to see if there's any other like-minded people who don't want their timelines filled with slop.

Hi HN!

I built NanoBibi.com, a free AI creation platform that lets anyone generate AI images, videos, and audio in one place.

The platform uses models like Google Nano Banana (Gemini 2.5 Flash Image) and ByteDance Seedance-1-Pro to create cinematic, high-quality content directly in your browser — no installs or accounts required.

This web application leverages multimodal LLMs (currently configured to use Gemini) to give business recommendations.

I want to do it for more sophisticated ML algorithms too, but first started with simple linear regression.

Hey HN!

I’ve been building ResumeAdapter, an AI-powered resume analyzer that helps job seekers understand why their resume isn’t getting interviews.

You upload your resume and a job description, and it instantly:

Highlights missing keywords and skills & and resumes Format gaps.

Gives an estimated ATS score (how well your resume matches the job)

Suggests improvements to increase interview chances

It’s designed for anyone tired of sending out 100+ applications with no replies. Would love your feedback on:

The UX (is it fast and intuitive?)

The accuracy of keyword detection

Anything that would make it more useful for developers and tech job seekers

Try it here: https://resumeadapter.com

Built solo, bootstrapped, and live! Happy to answer questions or share the tech stack if you’re curious.

Hey HN,

My friend and I built a Chrome extension called Pigeon because we kept Googling “Reddit + [product name]” whenever we wanted to know what real people thought about something.

Pigeon saves that step by looking at what’s on the page, finding related Reddit discussions, and showing the most helpful comments and opinions in one place.

We’ve both found it super useful when researching products or topics online and thought others might too.

Check it out and let me know what you think!