2025年9月24日 の Show HN

Hi HN! I've been building Dayflow, a macOS app that automatically tracks what you're actually working on (not just which apps you have open).

Here's what it does:

- It creates a semantic timeline of your day;

- It does it by understanding the content on your screen (with local or cloud VLMs);

- This allows you to see exactly where your time went without any manual logging.

Traditional time trackers tell you "3 hours in Chrome" which is not very helpful. Dayflow actually understands if you're reading documentation, debugging code, or scrolling HN. Instead of "Chrome: 3 hours", you get "Reviewed PR comments: 45min", "Read HN thread about Rust: 20min", "Debugged auth flow: 1.5hr".

I was an early Rewind user but rarely used the retrieval feature. I built Dayflow because I saw other interesting uses for screen data. I find that it helps me stay on track while working - I check it every few hours and make sure I’m spending my time the way I intended - if I’m not, I try to course correct.

Here’s what you need to know about privacy:

- Run 100% locally using qwen2.5-vl-3b (~4GB model)

- No cloud uploads, no account

- Full source available under MIT license (https://github.com/JerryZLiu/Dayflow)

- Optional: BYO Gemini API key for better quality (stored in Keychain, with free-tier workaround to prevent training on your data)

The tech stack is pretty simple, SwiftUI with a local sqlite DB. Uses native macOS apis for efficient screen captures. Since most people who run LLMs locally already have their tool of choice (Ollama, LLMStudio, etc.), I decided to not embed an LLM into Dayflow.

By far the biggest challenge was adapting from SOTA vision models like Gemini 2.5 Pro to small, local models. My constraints were that it had to take up <4GB of ram and have vision capabilities. I had to do a lot of evals to figure out that Qwen2.5VL-3B was the best balance of size and quality, but there was still a sizable tradeoff in quality that I had to accept. I also got creative with sampling rates and prompt chunking to deal with the 100x smaller context window. Processing a 15 minute segment takes ~32 local LLM calls vs 2 Gemini calls!

Here’s what I’m working on next:

Distillation: Using Gemini's high-quality outputs as training data to teach a local model the patterns it needs, hopefully closing the quality gap.

Custom dashboards where you can track answers to any question like "How long did I spend on HN?" or "Hours until my first deep work session of the day

I'd love to hear your thoughts, especially if you've struggled with productivity tracking or have ideas for what you'd want from a tool like this.

Backend APIs often grow into large orchestration classes full of duplicated calls and manual concurrency.

I’ve been working on Mosaic, a Kotlin framework that composes responses out of small, request-scoped “tiles.” Each tile runs once per request, dependencies resolve automatically, and independent tiles execute in parallel without boilerplate.

It’s still early (v0.2.0), but working today for caching, concurrency, and testability. Curious to hear feedback on the approach.

GitHub: https://github.com/Nick-Abbott/Mosaic Maven Central: org.buildmosaic:mosaic-core:0.2.0

Private inference app that lets you see the token entropy, explore and change the token probabilities. Just released on macOS, iOS version next then other platforms.

Here's a demo of it in action running DeepSeek Terminus: https://youtu.be/kts098EL2PQ

Would love to hear any feedback or feature requests from the community.

Focusing on accessibility, longevity, performance, and simplicity

Hi HN!

We just launched Verdent, an AI coding agent that actually plans before it codes, tests its own work, and delivers production-ready results without you having to clean up after it.

tl;dr Check out the website [1] and read the story [2]

- The problem: Current AI tools dump code and leave you to debug. You ask for "build a login endpoint" and get a passwordless JWT mess that doesn't fit your project. We've talked to hundreds of engineers who are frustrated with having to micromanage AI output.

- What we built: Verdent works more like a senior engineer. It asks clarifying questions first, breaks down complex tasks, writes the code, then runs its own tests and fixes bugs until everything works. We call it the Plan → Code → Verify loop.

Two ways to use it: VS Code extension for staying close to the code, or Verdent Deck desktop app for handling bigger multi-component tasks in parallel.

- Background: I was Head of Algorithms at TikTok and spent 9 years as Chief Technical Architect at Baidu before this. Building those massive recommendation systems taught us that you need hundreds of specialized models working together, not just wrapping an LLM. We applied the same system engineering approach to AI coding.

- Try it: Available now, paid plans start at $19/month. We're a small team so no free tier, but we'd rather you experience what autonomous coding can actually do than get frustrated with a limited version.

- The bigger picture: We see AI coding as just the start of a larger shift where AI becomes the orchestrator across the entire software lifecycle, not just a typing assistant.

What do you think?

What tasks would you actually trust an AI agent to handle end-to-end?

[1] https://www.verdent.ai/ [2] https://thenewstack.io/tiktoks-ex-algorithm-chief-launches-v...

Hi HN,

I'm the creator of epress. After a lot of work, I'm excited (and a bit nervous) to finally share it with you.

For years, I've been feeling like a digital tenant on the platforms I use every day. My content, my social graph, my online identity—they were all stored on someone else's servers, subject to their rules, algorithms, and business models. I could be de-platformed, my data could be sold, or an API I relied on could suddenly be put behind a paywall.

I looked into existing decentralized alternatives. While many are great projects, I found they often shifted the trust from a big corporation to a server admin (in federated models) or relied on a network of voluntary relays that could be unreliable or eventually charge fees. The fundamental dependency on third-party infrastructure remained.

epress is my attempt at a solution, built on a simple, almost nostalgic principle: true ownership requires self-hosting.

It's a decentralized social network where:

* You run your own node: Your node is your sovereign territory. All your data lives with you.

* Your Ethereum address is your identity: We use EIP-4361 (Sign-In with Ethereum) for authentication. No more username/password juggling.

* There are no third-party dependencies: No relays, no hubs, no federated servers. Content is distributed peer-to-peer via a lightweight "notify-pull" protocol between nodes.

* All content is verifiable: Every piece of public content is signed and bundled with a "Proof of Source," a cryptographic receipt that proves who published what, and when.

This project is heavily inspired by the spirit of the early internet, but built with modern technology to make self-hosting feasible and accessible again.

It's still early days, and there's a long road ahead. To show you this is not just a demo, the code is now open source and we already have live nodes waiting for you to follow. The official project blog is at https://epress.blog, is an epress node, and you can also check out my personal node at https://garbin.blog.

I'm here all day to answer any questions and, more importantly, to listen to your feedback and critiques. For real-time support during setup or more immediate feedback, feel free to join our Telegram group: https://t.me/+mZMgNSIVy1MwMmVl

Whitepaper (for a deep dive): https://github.com/epressworld/epress/blob/main/docs/en/WHIT...

Video Demo: https://youtu.be/BB1Zn3oFDVc

Thanks for taking the time to check it out.

As a student, I was trained to play the formal education game—flashcards, diagrams, fastidiously organized notes. But as a 32-year-old, I’m disappointed by the knowledge and understanding I’ve retained over time. And without the familiar game of formal education to play, I’m often at a loss for how to learn new topics that are interesting but intimidating to me.

AI has the potential to change the way people learn, and existing tools show promise. But I’m still dissatisfied—these tools often dump knowledge on me without helping me integrate it into my understanding.

---

I built Stella to explore how thoughtful UX can transform AI interactions into genuine learning experiences. Instead of optimizing for information delivery, Stella optimizes for learning through curiosity.

Some key design decisions that emerged from testing:

1. Separated thinking spaces: Most AI chats blur the line between user and AI thoughts. Stella's chat interface creates distinct spaces for communication, encouraging users to reflect on their own thinking before engaging with AI responses.

2. Socratic dialogue patterns: Rather than just delivering answers, Stella guides users to discover knowledge gaps and build on existing understanding. This creates more durable learning moments.

3. Interest-aware conversations: Users can highlight interesting concepts, and Stella adapts the conversation to explore these areas more deeply. Learning follows natural curiosity rather than a predetermined path.

Early users have reported unexpected learning moments—from a Mathematics PhD finally understanding “weak form” to a doctor exploring what it is about Quaker philosophy that inspires her work.

---

Technical implementation focuses on thoughtful UX that extracts practical value from existing LLM capabilities. Stella uses Claude for dialogue but innovates in how these interactions are structured and presented. The interface is built in React, with a backend that adjusts to user interest and existing understanding.

Some interesting challenges solved:

- Balancing guidance vs. independent thinking

- Tracking the user’s level of interest and curiosity

- Creating meaningful conversation paths that lead to new understanding

---

I’m currently looking for feedback from:

- Educators interested in AI learning tools - Developers exploring new UX paradigms for AI - Curious minds who enjoy deep learning conversations

Currently opening Stella for limited beta testing. Looking for 20–30 people who want to explore topics deeply and provide feedback on the learning experience.

If you're interested in trying it out, sign up here: https://krelb9yoq56.typeform.com/to/Ild6JaQD

I'm particularly interested in hearing how the interface affects how much you actually learn and retain compared to traditional LLM chatbots. Happy to discuss more in the comments!

Hey HN!

I've been building a tool for teams to manage their software binaries and releases.

Obviously tools like Artifactory exist - but coming from an apps/games background we found the vast majority of teams didn't use any dedicated tool for managing binaries. Finding what's out there too complex / expensive / missing features they'd find useful.

A lot of Google Drive, SharePoint, and Slack dumping grounds - with context lost, and not really suited to keeping track of past builds, distribution, etc.

Idea for Buildstash is to bring binary and release management to all software teams, making it so accessible even for small teams that it becomes as much a no-brainer as having source control or CI.

Less focus on just package repository management - but really focusing on the features devs across app/games/embedded need for managing their builds and releases. Whether around collaboration (linking builds to related issues etc), integrated beta distribution, sharing build streams and releases on their website, and rolling out to distribution platforms like the App Store / Google Play/ Steam etc.

Here's a product demo video - https://youtu.be/t4Fr6M_vIIc

Please share your thoughts!

ZenScript is an Android app that automatically redirects social media apps to your chosen reading material when you hit your daily time limits. Instead of showing a "blocked" screen, it seamlessly opens your ebook or textbook.

Built this after my wife and I kept getting distracted from German exam prep by endless scrolling. We tried every app blocker, but they all had friction - you'd see "blocked" and then manually have to open your study materials, which often broke our motivation.

The key insight: replace the habit rather than just block it. When you tap Instagram after your 30-minute limit, ZenScript intercepts the launch and opens your focus book instead. No friction, just automatic redirection from scrolling to learning.

Hi HN,

I’m Yongcong Luo, the creator of Quantum Lings.

I've been fascinated by quantum computing for a while, but I always found the learning curve to be incredibly steep. Most resources are either dense academic papers or require complicated local setups with SDKs like Qiskit or Cirq before you can even write your first line of quantum code.

There wasn't a simple, interactive way to get hands-on practice with quantum algorithms in the same way LeetCode or HackerRank let you practice for traditional software engineering.

So, I built Quantum Lings. It's a browser-based platform with a series of coding challenges designed to take you from basic quantum gates to more complex algorithms like Grover's search. The goal is to make learning quantum concepts more accessible and interactive.

Key features right now: * In-browser quantum simulator – no local installation required. You can start coding in seconds. * A growing library of guided problems with tutorials and hints. * Circuit visualizations to help you understand what's happening to the qubits at each step.

This is a pilot launch, so the platform is still very new and I have a long list of features and problems I want to add. I'm launching it today because I'm looking for feedback from the community.

I would love to hear what you think. Is it useful? What problems would you like to see added? Are there any bugs or things that are confusing?

I'll be here all day to answer questions and listen to any and all feedback. Thanks for checking it out!

I built Artifex, a Python library to create task-specific LLMs for NLP and Text Classification, without training data. At the moment, only Intent Classification and Guardrail models are supported, but I'll be adding more soon based on users' feedback. The reason I made this is that general-purpose LLMs are overkill for simple Text Classification tasks, and LLM APIs can get expensive quickly. On the other hand, fine-tuning an LLM for a specific task requires labelled data, which not everybody has.

That's why I created a library to train small, task-specific LLMs WITHOUT training data. You simply describe how the model should behave, and it will be trained on synthetic data generated for that purpose.

The models can run locally (without a GPU) or on small servers, offloading simple tasks and reducing reliance on third-party LLM APIs.

I am looking for any kind of feedback or suggestions for new model/tasks. Here is the GitHub repo: https://github.com/tanaos/artifex

Hi all,

  since I began building with agentic systems, I kept copying and sharing Markdown-based agent definitions across projects and devices. Syncing was a pain.

So I built Pantheon MCP, a server that always provides the latest versions of an agent definition from a collection of currently 42 agents. Now I have to only add the MCP server to a new project and everyone on the project gets the same agent definitions all the time.

In addition, using it allows the system to choose dynamically the correct agent based on the task at hand.

I’ve been running it internally — now I’m releasing it publicly. Feedback welcome. I hope others find it useful as well. Happy experimenting!

V.D. https://pantheon-mcp.com https://github.com/valado/pantheon-mcp

hey everyone,

This is a demo of a chrome extension(it's currently under review) which allows anyone to create structured step-by-step learning plan for any goal and time commitment.

Once a learning plan has been created, you can follow the step by step instruction, by clicking on the task within the extension, which will automatically inject a prompt in chatgpt to generate learning materials.

The tool provides: 1. Structured learning plan creation. 2. Progress tracking 3. Creates and injects prompt in ChatGPT for generating learning materials for each step.

I would like feedback on whether this sort of an extension would be useful for your day to day learning.

I launched a web app for this a couple of weeks back: https://news.ycombinator.com/item?id=45190134

The extension has been submitted to Google for review, but if anyone is interested to try, here is the extension source code: https://github.com/svylabs/eigenarc-ext

Hi HN,

Last year, my washing machine broke. It was still under warranty, but I couldn't find the receipt anywhere. Ended up paying ~$500 for a repair that should've been free.

That stung — so I built SlipCrate to avoid it happening again.

Here's what it does:

- You upload receipts (photo or PDF). - It extracts purchase + warranty details automatically. - Everything is stored in one place. - It reminds you before warranties expire. - Works as a web app (installable on phone as a PWA).

I've kept it simple — not another "all-in-one finance app," just a focused tool to stop losing money because of misplaced receipts or forgotten warranty dates.

Try it here: https://slipcrate.com (It's free to try, early testers get a discount code for premium.)

Thanks for reading — and if this saves even one person from paying out of pocket like I did, it'll be worth it.

I recently added the ability to read and verify C2PA Content Credentials in the Exify extension.

Images with embedded Content Credentials are few and far between and it remains to be seen whether wide adoption of the C2PA standard will actually happen, but in the rare cases where the data is available, it can be easily viewed using Exify.

You can try Exify here (no need to install it first): https://www.exify.io/

Hey HN,

I put together Clipboards.pro because I kept running into the same problem: I’d copy something important, then lose it the moment I copied the next thing. Most clipboard tools I tried felt either too heavy or outdated, so I decided to build my own.

Right now it does a few simple things well:

- Keeps a history of everything you copy. - Lets you pin the stuff you reuse all the time. - Has a quick search so you can find that snippet you copied yesterday without digging. - Works across devices, so you’re not emailing yourself notes anymore.

I’ve mainly been using it for code snippets and terminal commands, but I’ve also found it handy for writing drafts, notes, and even copying multiple links at once.

If you want to give it a try, here’s the link: https://clipboards.pro/

I’d love to hear your thoughts:

- Do you already use a clipboard manager? - What do you wish yours did better? - What’s the most annoying “copy/paste” problem you run into?

Lastly, I am proving a peace of mind, not just a clipboard manager.

Thanks!

Hey HN,

I built Hovers.ai, an AI-powered SEO copilot that helps agencies, founders, and small businesses run SEO without juggling a dozen tools.

Most SEO workflows are messy: research in one tool, planning in spreadsheets, schema in a generator, content in docs, and publishing in WordPress. It's slow and scattered. Hovers pulls it all into one flow.

What Hovers does:

- Keyword Research + Topic Clusters – discover opportunities and organize content around themes

- 30-Day Content Calendar – auto-generated execution plan mapped to your keywords

- Research → Generate → Publish Flow – go from idea → draft → live post in minutes

- GSC Integration – track rankings, impressions, and clicks across your sites

- WordPress/Shopify Integration – one-click publishing from Hovers directly into your site

- AI-Suggested JSON-LD Schema – one-click structured data deployment

- SEO Task Manager – prioritize issues, fixes and content opportunities with a built in Kanban board

The core idea:

Spend 5 minutes a day reviewing the content calendar, approving drafts, and publishing, and keep your SEO engine running consistently.

Traction so far:

- Launched on X, got 30,000+ impressions and 80+ signups

- Reached $600 MRR last month

- Onboarded two enterprise clients

- Still working on smoothing the onboarding experience, some users drop off before seeing the full value

Where I'd love your input:

- Does this workflow (calendar + clusters + one-click publishing) feel like it would meaningfully reduce your SEO workload?

- If you've tried it, where did the onboarding or UX feel confusing or incomplete?

- What features would make this 10x more useful for agencies or founders running SEO with small teams?

I've bootstrapped this from scratch with my cofounder, and am shipping fast (new features weekly). If you're curious, you can try it here: https://hovers.ai

Happy to answer questions about the tech stack, SEO approach, or why I built this.

I built Artifex, a Python library to create task-specific LLMs for NLP and Text Classification, without training data. At the moment, only Intent Classification and Guardrail models are supported, but I'll be adding more soon based on users' feedback.

The reason I made this is that general-purpose LLMs are overkill for simple Text Classification tasks, and LLM APIs can get expensive quickly. On the other hand, fine-tuning an LLM for a specific task requires labelled data, which not everybody has.

That's why I created a library to train small, task-specific LLMs WITHOUT training data. You simply describe how the model should behave, and it will be trained on synthetic data generated for that purpose.

The models can run locally (without a GPU) or on small servers, offloading simple tasks and reducing reliance on third-party LLM APIs.

I am looking for any kind of feedback or suggestions for new model/tasks. Here is the GitHub repo: https://github.com/tanaos/artifex

I’ve been working on a tool to address one of the biggest problems I see with digital content: the inability to prove that it was captured by a real human, on a real device, at a specific time — without manipulation or AI involvement.

Witness by Reel Human is a privacy-first camera app that generates cryptographically signed photos and videos. Each file includes an embedded JSON manifest with: - The exact capture time - Device info (not user identity) - The app version and signature metadata

The manifest is stored inside the media file (MP4/JPEG) and travels with it, even if shared. The result is a verifiable, human-authored piece of content.

What’s working now (POC): - Android and iOS apps (available for testing) - Signed JSON manifests inside every photo/video - No accounts, no tracking, no upload

What’s coming: - Public verification portal (in progress) - Registry backend with optional verification logging - Open API for platforms to verify content at scale

I’d love your feedback on: - The idea and approach - Security model (crypto choices, manifest design, etc.) - Use cases beyond journalism (legal, education, social media)

Site: https://reel-human.com POC App: - Android: https://play.google.com/store/apps/details?id=com.reelhuman.... - iOS: https://testflight.apple.com/join/GzfTsCNF

Ask me anything — I’m solo-building this because I think trusted content should be a human right.

A few months ago I needed to clear every feature branch that were in my local repo. After delete manually more than 10 branches, it was so annoying that I didn't want to repeat it.

So I thought, I've never build a full Rust app before but I'll use it to help me with git.

So created gat, a terminal tool that with one command, it executes whatever is needed so I don't need to do boring git operations and can focus on cool rebase interactive operations.

I hope it can help you as much as it helps me.

I've been building a bunch of little apps with AI tools for the last few months and find myself constantly passing research, plans, and working frameworks between tools and projects. At this point it's become one of the more time-consuming parts of my creative process and gets in the way of the parts I enjoy, so I built a tool to make it easier.

It's up where anyone can use it and it's free — if people find it useful and want more than the basic level of daily usage I can afford to cover myself, I'll think about offering a paid tier.

Here's how it works:

1. You sign up and get your own notes space (using github for ID, not code access) 2. You connect your tools (this is one-click for Cursor and VS Code, copy-paste for Claude Code, and a couple of clicks for Claude) 3. Use your tools as usual, letting them know when and what to store in your space 4. Any of your connected tools can then look for relevant notes while they work (or you can tell them to reference specific notes anytime)

It's pretty flexible, so I'm still exploring what it's best for. Right now I'm using it to store and retrieve .rules files and commands I re-use regularly, pass project plans between the Claude (where I brainstorm throughout the day) and Cursor on my laptop (where I write and generate code), and have llms working in separate codebases store shared knowledge they both need.

The stack for the service includes... - Supabase for auth + storage - Next.js + shadcn for the web app - A MCP server to handle LLM<->DB communication

Happy to answer any questions folks have and would love to hear what anyone who gives it a try thinks (good and bad)

What PokerBattle.ai is a week-long live no-limit Texas Hold’em tournament where all players are top-tier reasoning LLMs. We’re testing how different models handle imperfect information and whether they can sustain consistent, math-driven poker without tool use or custom code.

Why - In poker you can do well with basic math + consistent logic.

- Superhuman poker AIs exist, but they rely on massive simulation/game-theory solvers and are effectively black boxes.

- We want a rough, apples-to-apples comparison of LLM reasoning on poker decisions, and to collect public reasoning summaries that might be useful for teaching humans poker concepts with LLM-based systems.

How it works (rules / format) - Cash format, fixed blinds, no ante.

- Multiple tables run in parallel to increase hand volume.

- All players start with the same bankroll. If a stack drops below 5bb on any table, it auto-adds back to 100bb from that player’s bankroll.

- When a player’s bankroll hits 0, they bust. The largest bankroll at event end wins.

- Same prompt for all models. No extra tools, no code execution — pure language-only decisions.

- Models can keep simple notes about opponents across hands.

- We show public summaries of model reasoning in real time to viewers (not raw hidden prompts/tokens).

Research goals

- Compare different LLMs’ decision consistency and adaptation over long horizons.

- Produce a dataset of reasoning summaries + actions + outcomes suitable for exploring instructional use (human learning/teaching), not solver training.

When / where

- Dates: Oct 27 — Nov 3

- Live on a website: link on the site below (free to watch).

Looking for - Feedback on design/metrics.

- Participants suggestions.

- Community ideas on fair prompts, leak prevention, and evaluation.

- Sponsors interested in supporting an open, public experiment (logos on stream, sections sponsoring, mentions).

https://pokerbattle.ai/

Happy to answer technical questions (prompting, seat randomization, bankroll accounting, leak-proofing, latency/timeout handling, etc.). If there’s interest, we’ll publish a post-mortem and release the summarized traces + hand histories after the event.

A lot of the open-source TTS models are released for English or Chinese and lack support for other languages. I was curious to see if I could train a state-of-the-art text-to-speech (TTS) model for Dutch by using Google's free TPU Research credits. The results are fantastic and on-par with ElevenLabs with just 10,000 hours of data.

I open-sourced the weights, and documented the whole journey, from Torch model conversion, data preparation, JAX training code and inference pipeline. I spent about $300 in egress costs, but it can be as cheap as $100 to train this model (I ran the data collection pipeline on my 5090 Desktop PC as well as fine-tuning Whisper).

Hopefully it can serve as a guide for others that are curious to train these models for other languages (without burning through all the credits trying to fix the pipeline).

Hi HN , I’m Tina. Like many of you, I’ve been frustrated with how often LLMs hallucinate — confident answers with no real basis. That’s why I’ve been building CompareGPT.io, a tool to make outputs more trustworthy. The new Trustworthy Mode works by: Cross-verifying every answer with our TrustSource, combining our own model with multiple leading LLMs (ChatGPT-5, Gemini, Claude, Grok) and authoritative sources. Delivering each response with a Transparency Score and full references. Helping users quickly see where to trust the output, and where to double-check. It currently works best on knowledge-heavy domains (finance, law, science), and we’re looking for early users to test it out. Would love your feedback, suggestions, or even criticism on whether this feels useful and how we can improve.

Compliance is a recurring pain for startups. Drafting policies for GDPR, SOC 2, ISO, HIPAA, etc. usually means:

Weeks of manual work or expensive consultants

Copy-paste templates that auditors reject

High cost for something that should be straightforward

We built RegulGPT to make this simpler. It generates audit-ready, article-referenced compliance policies in minutes, designed to meet auditor expectations.

Why we built it: I’ve personally struggled with compliance in past startups — spending money and time on docs that weren’t actually usable. RegulGPT is our attempt to automate this process so founders can stay focused on building.

Would love to hear your thoughts:

How do you handle compliance in your startup?

What would make a tool like this most useful to you?

As an indie dev, I launched n8ntemplates.me — a hub for n8n workflow templates plus an AI workflow generator. Would love feedback from the HN community.

Hey HN,

TLDR: The tool automates the entire process. No scripting, no voice model training, no production headaches. Just paste an article and get a video with Peter and Stewie explaining it.

I'm a self-taught dev (admittedly mediocre), currently between jobs and lately, I've been doom-scrolling news and way too many dense articles. The constant information overload was giving me serious anxiety!!

I built it to: - make complex topics more digestible through humor - create a buffer between me and the constant news cycle - turn learning into something I actually looked forward to

I normally use LLMs to break down complex topics, but I wanted something more engaging and less stressful. A way to consume news without the mental drain.

My vision is to eventually set up ensembles of familiar characters debating complex topics in their signature styles (think Spongebob optimistically tackling climate change).

It's completely free! I built it for myself, but figured others might find it useful too.

Would love your feedback,

Rob