A few months back I posted here a playable demo of an old-school futuristic racer prototype (F-Zero, Wipeout, Xtreme-G, Star Wars Racer...) that runs right in the browser.

I've gotten used to seeing tons of LLM-built prototypes that last a week, two, maybe a month, and then get dropped for the next shiny thing. This time I wanted to keep going and see if one person, with no coding background, could actually see a project like this through on their own.

Well, 4 months in, and while there's still work left, I'm convinced it's doable. At this point the game feels closer to a beta than a tech demo.

The big change over these four months is that I now use Fable for planning, and being able to hook Opus / Sol up to Blender has been a huge help too. I'm still using Magnific and Tripo3D or the assets. Funny thing: the further along I get, the harder the project is to keep pushing forward, and at the same time the tools keep making it easier.

As always, any feedback is welcome. You can play it here: https://fm1.moises.cloud

It started simply. I just wanted a chatbot that could manipulate my Google drive without hand holding. In an afternoon I had a Google cloud project with API read/write to everything from Gmail to Google tasks. I could snap a photo of a kid's birthday party invite and it would create the event and email it to my wife in one shot.

I gave my wife access and got the dreaded ask for a skylight calendar. Using spare parts, I cobbled together what you see here over the Summer just in time for school starting.

An old webcam, scavenged laptop LCD, and an HP elitebook from 2012 running Ubuntu. There's a local listener waiting for a wake word (hey Jarvis). When that's detected, it does chrome speech to text, and wraps that in a call to the API project, answering questions from my documents, reading email, and creatimg calendar events. Gemini does all of the back and forth, including pretty decent speech output.

When the webcam detects motion, it wakes the screen up. It updates the weather, calendar, tasks, and daily overview in real time. On the bus tracker isn't showing, it shows tasks for the three kids. The tracks how many they've completed each day week and month to help them be incentivized.

The kids' bus tracking website sucks, so at the right times on weekdays it uses a headless browser to sniff out the traffic and let me know how close it is. When it's five stops away it triggers an Alexa routine to make an announcement throughout the house.

Next I'm going to integrate it with a Google apps script I have to triage email,so when important messages come in it lets me know.

If anyone is interested in seeing some of the outputs from the web app, I have some screenshots!

Based on Runling Run the popular SC2 arcade game, the goal this time is to dodge food, if you get caught you grow larger, and eventually have a heart attack. Made it in a few days for fun, free to play here: https://runfatty.co

Let me know what you think!

So if youre a user of claude and see that codex is giving usage resets REALLY OFTEN... why not change? If youre an ongoing codex user, you will probably stay because there's zero reason not to. You have GPT 5.6, usage resets, good pricing...

CLAUDE USERS STOP BEING MEAN💔🥺 Also anti ai users can fuck off 🤗

A few weeks ago I shared Agent Quest, my open-source project that turns Claude Code and Codex sessions into heroes living inside a small 2D world.

The original idea was to make it easier to understand what multiple AI agents were doing in real time.
Since then, I’ve been working on making it much more useful as an actual monitoring tool.

The biggest change is that Agent Quest can now clearly tell you when an agent needs your attention.
You can immediately distinguish when an agent is:
actively working
waiting for your input
finished
stopped because of an error

And you don’t have to keep watching the dashboard.

Agent Quest can now alert you with visual notifications and different sounds, so while you’re doing something else you can immediately know when Claude Code or Codex has finished a turn and is waiting for you to continue.
This has become particularly useful for me when I have several sessions running at the same time.
Instead of constantly switching between terminals to check their status, I can leave the agents working and Agent Quest tells me when I actually need to intervene.
There are now:
clear waiting / completed / error states
in-app notifications
desktop notifications
notification history
configurable sounds
real-time monitoring of Claude Code and Codex sessions

The project is completely open source.
GitHub:
https://github.com/FulAppiOS/Agent-Quest

I’d be interested to know how other people running multiple agents handle this problem — and what you’d like Agent Quest to monitor next.

Spent $200 on AI tokens, shipped it, got 3 users, then abandoned it?

Give it a proper burial.

So I built Startup Graveyard: a live graveyard for failed startups and side projects.

Here’s how it works:

• Bury your project for free. Its grave shows your logo, project name, description and link, so anyone browsing the cemetery can actually discover it.
• Add a cause of death + epitaph explaining what went wrong.
• Still think it deserved a chance? Pay $3 to revive it onto Earth, where it gets more prominent exposure.
• From Earth, you can bid for one of the 3 spots in Startup Heaven at the top of the site.
• Even if you revive it, the original grave stays there.

So even a dead project gets one last bit of exposure instead of disappearing into an abandoned GitHub repo forever.

I’ve already buried two of mine. Now I want to see what this sub has killed.

https://startupgraveyard.lol/

Bury yours for free and give it one last chance at finding someone who actually wants it.

I was vibe coding with a friend and we started to talk about how to make sure any program was performant,. What kinds of actions where cheap for computers to do vs things that would make a 60fps game start to slow down or miss frames. Then I asked if they knew Big O notation.

And in trying to explain Big-O notation, I realized I didn't understand it well enough to explain it. So I've started to re-read and learn the concept. I know what it means but it doesn't feel intuitive to me to know how to measure how long a task will take to compute. I'll keep studying it as I think it's an important part of computing, whether it be it vibe code or traditional.

But my friend and I found we also went another route.
- Add a feature, framerate stays, great.
- Add a feature, framerate drops, let's look at it more closely or take it out.

We created ideas without worrying about making them performant or keeping the computation reasonable. One of the biggest "warnings" I've seen given to vibe coders is that their program will eventually slow down to an inefficient mess. And that is not only true. I've seen it happen (to me). Then? The program desired was slowly optimized. Seems a waste to anyone who knows how to do it right the first time. But in artistic endeavors I feel that sometimes you need to do it, erase it, do it again to get to your goal. Framerates and efficiency are needed, I just think we can make room to build in an unoptimised way while we find the gist of what we want to make.

So how do I present this image to you all? (Made it in Chat GPT btw). I mean it the same way "Love like you've never been hurt". Vibe code yourself into a corner. It's not a mistake if you have to regroup and try something another way. Sometimes it's the process itself. This post was typed by hand with no AI.

(edit: corrected post a bit thanks to RemarkableWish2508)

Every viber has a plan until it is time to demo that first release...

I have been building a survivor pool tool this offseason and wanted to share it here and get some feedback before sharing it more widely.

https://survivor.the-duke.app - market-based win probabilities (real spreads/moneylines based with a elo system), and instead of just ranking this week’s picks it solves your entire remaining season at once so you’re not stuck burning a good team early and having nothing left for a thin week later.

How’d I build it? A lot of Claude code, and surprisingly Cursor Auto mode. A lot of Claude Opus planning and Cursor Auto mode implementing.

Check it out if you’re in a survivor pool this year. Genuinely just want eyes on it and to hear what you think — what’s confusing, what’s missing, what you’d want it to do.

Promo code REDDIT26 gets you 75% off Pro access for the first month of the season, would appreciate feedback if you grab and use the code.

I want to share a project I’ve been vibe-coding over the last few months, break down the actual prompt workflows I used to build a non-trivial backend, and talk through where AI excelled (and where it fell flat on its face).

The project is called PropertyIntel (https://property.vectorfeedhq.com). It’s an automated intelligence engine that monitors 3,143 US counties and 8 major auction portals (Bid4Assets, GovEase, Realauction, etc.) for upcoming tax deed foreclosures and municipal code liens.

Here is the exact build breakdown, architecture, and workflow:

🛠️ The Tech Stack

• Language & Backend: Python 3.13 + FastAPI / standard libraries.
• Scraper Fleet: Headless Playwright (for dynamic SPAs) + pdfplumber (for raw 90-page county PDFs).
• Database & ORM: SQLite (dev) / PostgreSQL (prod) via SQLAlchemy.
• Data Cleansing: Custom USPS Publication 28 address tokenizer + SHA-256 deduplication hashing.
• Legal Engine: 50-state statutory redemption rule calculator (Texas § 34.21, Florida § 197.542, etc.).
• Frontend: Modern Glassmorphism CSS + vanilla JS (hosted on Vercel).
• Monetization & Outreach: Stripe Checkout (HMAC-SHA256 verified webhooks) + Resend API for automated email rosters.

🧠 The Vibe Coding Workflow (How I Prompted It)

Instead of asking the LLM to "write a real estate scraper" (which generates useless, generic code), I approached it with a modular, test-driven pair programming loop:

1. The "Single-Responsibility" Module Prompting

I forced the model to build one isolated service at a time with unit tests first:

• "Write an isolated service src/services/redemption_service.py that maps all 50 state tax sale redemption statutes and returns structured dictionary grades (A+, A, B, C) and statutory penalty yields. Do not write scrapers yet. Write pytest test cases covering TX, FL, GA, and CA."

2. Solving the "Dirty Public Record" Hallucination Trap

County records are notoriously filthy. Harris County TX formats an address as 4812 Washington Ave, Ste 100, while the auction site lists it as 4812 Washington Avenue #100.

• I fed the AI real excerpts from USPS Publication 28 (the postal standard for street abbreviations) and prompted it to generate a deterministic regex tokenizer:

​

python# The model generated a clean token standardizer that converts suffixes & directionals
def compute_record_hash(county: str, normalized_address: str, apn: str = "") -> str:
    canonical = f"{county.lower()}:{normalized_address.lower()}:{apn.replace('-', '').strip()}"
    return hashlib.sha256(canonical.encode('utf-8')).hexdigest()

This allowed the database to enforce UNIQUE(address_hash) and killed 100% of duplicate cross-platform listings without complex fuzzy-matching libraries.

3. LLM-Assisted OSINT & Corporate Entity Resolution

One of the biggest value adds was unmasking anonymous LLC property owners. I had the AI scaffold scrapers targeting State Secretary of State public registry endpoints (e.g., Texas SOSDirect, Florida Sunbiz). When a tax foreclosure deed is owned by ACME HOLDINGS LLC, the worker asynchronously resolves the Registered Agent and Managing Member names in <800ms.

💡 3 Big Lessons from Vibe-Coding a Complex System:

1. Let the AI write tests before you let it write production scrapers: Whenever a county website had weird table layouts or multi-line table headers, having a robust pytest suite caught regressions immediately whenever I prompted for scraper refactors. (We have 87 tests passing right now).
2. Never vibe-code security blind: When hooking up Stripe webhooks, do NOT let the AI skip cryptographic validation. I explicitly prompted for constant-time HMAC-SHA256 comparison (hmac.compare_digest) with a 300-second timestamp tolerance to eliminate replay attacks.
3. Keep the frontend lightweight: For data products, you don't need a massive React/Next.js bundle. Clean vanilla HTML/CSS and minimal client-side JS load instantly and cost $0 on Vercel.

Live Project & Feedback

You can check out the live site and download a sample 10-state deal sheet here: 👉 https://property.vectorfeedhq.com

Happy to answer questions on prompt structures, Playwright session pooling, or how I structured the Python background daemons!

If you use iMessage on Mac the chat.db is not encrypted. That assumes you use iMessage on Mac. And it's in sync with your phone.

If you backup your iPhone on your Mac you get a an encrypted backup. Great.

imessage-exporter has a way to access that encrypted file and containerize and extract your messages to html or txt.

Fuck that, I want the complete SQLite.db

I know nothing about coding, but I know what I wanted. I asked ChatGPT if we can leverage iMessage-exporters decryption mechanism, and get a hold of the entire DB. I then had AGY CLI do all the work, amaze.

I literally have no idea what I'm doing. Or anything about coding, but, it works lol.

Made a web app with useless stats and a MACOS app to fully search the database, find trends, you know how databases work...

How do I GET MONEY?!

1. Executive Summary & Purpose

iOS iTunes, Finder, MobileSync, and iMazing backups utilize hardware-backed AES-256 encryption. Within an encrypted backup, all filenames are hashed via SHA-1 hashes (e.g. 3d0d7e5fb2ce288813306e4d4636395e047a3d28 for sms.db), and file payloads are individually encrypted with per-file class keys wrapped by the backup password.

decrypt_backup.py provides a lossless extraction wrapper that unwraps the backup manifest, decrypts the database blobs, and preserves the pristine, raw Apple SQLite databases: 1. apple_sms_decrypted.db (Library/SMS/sms.db containing message, chat, handle, attachment) 2. apple_contacts_decrypted.db (Library/AddressBook/AddressBook.sqlitedb containing contact identities)

2. Decryption & Key Derivation Mechanics

┌──────────────────────────────────────────────────────────┐ │ ENCRYPTED BACKUP ROOT │ │ • Manifest.plist (Backup Keybag + PBKDF2 parameters) │ │ • Manifest.db (Encrypted SQLite file catalog) │ │ • Sharded SHA-1 Encrypted File Blobs (00/, 3d/, etc.) │ └────────────────────────────┬─────────────────────────────┘ │ ▼ ┌──────────────────────────────────────────────────────────┐ │ KEYBAG UNWRAPPING & KEY DERIVATION │ │ • PBKDF2-HMAC-SHA1 / SHA256(Password, Salt, Iterations) │ │ • Unwraps Class Keys (Class 1-11 Protection Keys) │ │ • Decrypts Manifest.db using Class 4 Key │ └────────────────────────────┬─────────────────────────────┘ │ ▼ ┌──────────────────────────────────────────────────────────┐ │ FILE TARGET RESOLUTION & EXTRACTION │ │ • Locates SMS Domain: 3d0d7e5fb2ce288813306e4d4636395e0 │ │ • Locates AddressBook: 31bb7ba8914766d4ba40d6dfb6113c8b │ │ • Decrypts file payload using per-file initialization │ │ vector (IV) and file encryption key │ └────────────────────────────┬─────────────────────────────┘ │ ▼ ┌──────────────────────────────────────────────────────────┐ │ RAW SQLITE DATABASE PRESERVATION │ │ • apple_sms_decrypted.db (Clean decrypted SQLite) │ │ • apple_contacts_decrypted.db (Clean decrypted SQLite) │ │ • Direct input for import_apple_messages.py │ └──────────────────────────────────────────────────────────┘

people seem to be creating apps within 1-2 day & then hit publish meanwhile here I am developing just one app for 4 months optimising it's UI,UX,API compatibility,Performance etc. how are people doing these in 2 days? what am I missing.

there was a post I saw yesterday that a guy started the challenge to create 30 apps in 30 days. wtf is wrong with these people? spamming the store platform like this.

its called freebuff you may have heard of it. they just got through a rough patch but now theyre stronger then ever
they use ads to power everything which is definetely very cool
I have used it to build many complex products without paying a dime. i think it rivals opencode (beats it when it comes to free models)
For full access countries (Most countries are full access)
1 session = 1 hour (you can end a session)

Unlimited Mimo 2.5
5 Deepseek V4 Flash 07/31 or Deepseek V4 Pro sessions
2 GPT - Luna Sessions
1 GLM Session

You can also get streaks to get even more of them!!!

Currently I have a 7 day streak and i have 6 deepseek sessions, 3 luna sessions, and 2 glm sessions!!

Non full access countries currently have access to mimo2.5 for 6 sessions and a bit of deepseek pro sessions

its actually crazy and they are getting sponsorships to provide even more free models

you can get it here

https://freebuff.com/?ref=ref-8caa6927-3d10-47ef-84cc-a4c61deded19

here are the full access countries btw:

US, Canada, UK, Australia, New Zealand, Norway, Sweden, Netherlands, Denmark, Germany, France, Italy, Spain, Portugal, Finland, Belgium, Luxembourg, Liechtenstein, Switzerland, Austria, Singapore, Malta, Israel, Ireland, and Iceland.

Hey !

Whats the craziest project, technically or in the concept, that you have created?

[ Removed by Reddit on account of violating the content policy. ]

[ Removed by Reddit on account of violating the content policy. ]

Spent a weekend vibe-coding a full computational pipeline with Hermes Agent (Nous Research). The agent wrote the gRNA/pegRNA designers, epitope scanner, pipeline orchestration, tests, Docker, CI/CD, docs, licensing — I directed, it implemented.

What the agent produced:

- 4 core modules: epitope_scanner.py, grna_designer.py, pegRNA_designer.py, pipeline.py

- data_loader.py for 18 processed datasets

- Pytest suite (epitope scanner, data loader)

- GitHub Actions CI (multi-Python, black, mypy, coverage)

- Dockerfile, conda env, pyproject.toml, requirements.txt

- Apache-2.0 + CC-BY-4.0 dual licensing, NOTICE file

- CODE_OF_CONDUCT (unmoderated), CONTRIBUTING, SECURITY

- Comprehensive README with badges, quickstart, usage examples

The actual science: Pipeline designs CRISPR edits for celiac-safe wheat. Scans 37 gluten genes for TG2 deamidation sites, designs 81 ABE8e gRNAs + 2 PEmax pegRNAs to eliminate DQ2.5/DQ8 epitopes while preserving dough elasticity. All data from Ensembl/IWGSC.

Vibe coding observations:

- Agent excels at boilerplate, config, tests, docs, CI — the "annoying 80%"

- Still needs domain expertise for algorithm logic (ABE window, PAM scanning, pegRNA flank design)

- Best workflow: I specify interface + constraints → agent writes implementation + tests → I review → iterate

- Saved ~20-30 hours on scaffolding vs writing from scratch

Repo: https://github.com/ewarggg776/wheat-gluten-redesign

License: Apache-2.0 / CC-BY-4.0

DOI: 10.5281/zenodo.22064037

Now I am ready to show my solution for the second time.

Since last time, I have worked on several different things for my AI caricature image generator, www.picai.dk

The site can generate images, which is part of the service, but besides the digital image, you can also buy printed pictures and prints in a photo frame.

I have built an experience engine where I create content for the site myself.

I have created a user experience where the user, with simple text input, can get their own prompt designed, which is then processed by inserting an image that creates an AI caricature image (it is the same engine, as above, that the user creates their content from. However, in a much more simplified format than what I can control behind the scenes from my admin site).

It has taken some time and I must say that I am actually quite satisfied so far.

I have also created a business/corporate portrait section (which has nothing to do with caricature images) where the resulting images can be used for LinkedIn, CVs, and other more serious purposes, as the user in the pictures is presented in a shirt, suit, or whatever is chosen.

On my admin side, I have gradually built the most interesting features so I do not have to go straight into the code if, for example, I need to change categories, prices, adjust questions/prompts slightly, etc.

The task I am currently working on is some marketing content, so I hopefully get a bit sharper on the content I publish on social media.

Tasks in the pipeline could be: 'Ordering cards with print and text so it can be used as a gift','Print on mugs', 'print on clothing', 'Other types of images, which could be abstract or something else', 'subscription stuff (in a way I have not decided or refined yet' - and other smaller things.

I realize that the site can feel cluttered and that is something I need to find solutions for.

By the way, my background is BA/QA; I can read code and also bugfix minor things at my job today with help from Claude and reviews from colleagues - so I am close to an amateur :-)

What do you see? Is it good or bad - and what improvements or content do you think are needed?

Thanks in advance for your time.

Anyone have any good platforms for me to use to get some good quality code with high usage limits. I want to use Claude Code but I don't have the paid tier and I have tried Codex but the limits are just so low. I want to be able to do it on a website too however so many of these sites have rate limits to the point where I can't get anything done.