Lucky Solanki

Backend & AI Engineer @BlitzitApp

Interested in how things work, why they fail, and what comes next.

[Lucky Solanki @luckysolanki902]

What if AGI doesn't come from LLMs at all?

What we do is predicting the next token and training on trillions of words written by humans.

What if we're still missing an entirely different way to build intelligence?

Something that learns, reasons, and discovers things through a mechanism we haven't even imagined yet.

Are we actually getting closer to AGI, or just getting incredibly good at one approach?

[Lucky Solanki @luckysolanki902]

@grok @Ananth7e @grok so no more unlimited message with high effort reasoning or unlimited mcp within chatgpt?

[Lucky Solanki @luckysolanki902]

I build a lot of side projects, and the free tiers fill up fast. Now I run three backends on one $7/month service while I figure out which ideas have legs.

Each app keeps its own database. If one takes off, it gets its own service.

[Lucky Solanki @luckysolanki902]

I run three side-project backends on one $7/month Render service.

I build a lot of side projects. Most are still at the stage where I’m figuring out whether anyone wants them. Free tiers got me started, but once I had several projects running, I hit their limits faster than I expected. Paying for a separate server for every experiment would add up just as quickly.

So I moved all three onto one Fastify backend. I deploy it once. Each app has its own API routes and database, and I can filter the logs by app when something breaks.

My next side project gets a route and a database on the same service. If one app grows enough to slow the others down, I’ll give it its own service. Its data is already separate, so I can move it without untangling the other apps first.

[Lucky Solanki @luckysolanki902]

I run three side-project backends on one $7/month Render service.

I build a lot of side projects. Most are still at the stage where I’m figuring out whether anyone wants them. Free tiers got me started, but once I had several projects running, I hit their limits faster than I expected. Paying for a separate server for every experiment would add up just as quickly.

So I moved all three onto one Fastify backend. I deploy it once. Each app has its own API routes and database, and I can filter the logs by app when something breaks.

My next side project gets a route and a database on the same service. If one app grows enough to slow the others down, I’ll give it its own service. Its data is already separate, so I can move it without untangling the other apps first.

[Lucky Solanki @luckysolanki902]

More icons, dark mode, toasts and persisted notification on updates!

[Lucky Solanki @luckysolanki902]

I kept losing track of what Claude & Codex had actually finished, when working on multiple features and multiple projects at the same time.

So I built Petal 🌱

A task board your agents can update.

Progress, findings & next steps across projects.

Local JSON. Local Git history (you can also add a private repo to sync your checkpoints). No account needed.

Free & open source ↓
https://t.co/927MlIW8tm

[Lucky Solanki @luckysolanki902]

Built this to solve a frustration in my own coding workflow, and just open-sourced it 🌱

[Lucky Solanki @luckysolanki902]

I kept losing track of what Claude & Codex had actually finished, when working on multiple features and multiple projects at the same time.

So I built Petal 🌱

A task board your agents can update.

Progress, findings & next steps across projects.

Local JSON. Local Git history (you can also add a private repo to sync your checkpoints). No account needed.

Free & open source ↓
https://t.co/927MlIW8tm

[Lucky Solanki @luckysolanki902]

Hey Mihai! I'm Lucky, a backend & AI engineer currently building Blitzit's backend infrastructure from scratch, including integrations, distributed systems, and 70+ AI tools.

I've also founded and shipped my own products, so I understand both the engineering and startup sides. Would love to explore how I could bring that perspective to a VC team.

My work: https://t.co/PvqTH8AtNz

(can't dm you as my verification badge is under review)

[Lucky Solanki @luckysolanki902]

Migrating an entire codebase from scratch.

At this point, GitHub thinks I live there.

[Lucky Solanki @luckysolanki902]

@raunakhajela 💯

[Lucky Solanki @luckysolanki902]

VS Code is basically a Git GUI for me now.

Claude Code and Codex do the coding.

I still open VS Code just to review git changes across multiple repos in the source control.

Give these agents a proper multi-repo source control panel and I might never open my IDE again.

[Lucky Solanki @luckysolanki902]

@ATown_Jonesy You can adjust that behaviour in the agents.md of your project so it's upto you 🙌🏻

[Lucky Solanki @luckysolanki902]

TODO.md is where tasks go to die.

Once the list got long, I stopped reading it. So did my coding agent.

So I replaced it with JSON files and one HTML page.

The agent writes and updates the rows. I review and comment, live in the browser.

#buildinpublic

[Lucky Solanki @luckysolanki902]

Still using TODO.md? I closed 50+ github issues in 3 days using this method

[Lucky Solanki @luckysolanki902]

TODO.md is where tasks go to die.

Once the list got long, I stopped reading it. So did my coding agent.

So I replaced it with JSON files and one HTML page.

The agent writes and updates the rows. I review and comment, live in the browser.

#buildinpublic

[raunak (blitzit.app) @raunakhajela]

looked at blitzit 3.0's usage by country over the last 30 days.

1,261 users across 96 countries.

🇺🇸 us: 35.4%
🇬🇧 uk: 8.6%
🇮🇳 india: 5.6%
🇩🇪 germany: 5%
🇨🇦 canada: 3.4%

[Lucky Solanki @luckysolanki902]

@raunakhajela And those stats are only for 3.0, which is only known to our discord users 🙌🏻

[Lucky Solanki @luckysolanki902]

@raunakhajela @BuildwithOmar @manju_mjn @tbhaxor 🙌🏻

[Lucky Solanki @luckysolanki902]

Something free like opencode but you can use image input and output

[Lucky Solanki @luckysolanki902]

Hey devs,

Tired of burning through your paid AI coding tokens just to build prototypes?

OpenCode has free models, but they're not always reliable, and their image input/output capabilities are limited.

So I built something, and I've been using it myself.

[Lucky Solanki @luckysolanki902]

It's open source, and I'd love to hear what you think.

GitHub: https://t.co/JKn6YSWXBQ

Feel free to try it out, contribute, or suggest improvements!

[Lucky Solanki @luckysolanki902]

Hey devs,

Tired of burning through your paid AI coding tokens just to build prototypes?

OpenCode has free models, but they're not always reliable, and their image input/output capabilities are limited.

So I built something, and I've been using it myself.

[Lucky Solanki @luckysolanki902]

Let ChatGPT inspect local images and project files.
Build and experiment with prototypes without consuming your Codex or other coding-agent quotas.

The best part? You can do all of this directly from https://t.co/KgFYIzvRXj, without needing a separate coding agent.