Devansh ✨

@arpit_bhayani What was the moment you realized building a control plane wasn't enough for you?

2 months ago, I was interviewing at Atlassian, and it was a place that I really wanted to join - because I was being interviewed for a team that was building a control plane for transactional databases ... so, of course :) Had one round of interview, and it went really well. I was sooo looking forward to interviewing further and cracking it, but Atlassian went into a hiring freeze. Was totally bummed out. Fast forward to today, here I am, getting invited for an AMA session :) What could have been an internal talk is now an external session - funny how things work out sometimes. Thanks a ton, Puneet Patwari (@system_monarch), for inviting me. It really means a lot. I hope everyone had a great time. I surely did. Also, thanks for asking such an awesome set of questions - they made me pause and really self-reflect. Didn't get the role, but still got the room - super grateful.

@mattpocockuk what's the map supposed to look like

I've found myself writing: "I don't know this area of code well. Go up a layer of abstraction. Give me a map of all the relevant modules and callers." Might need a new skill here. What should I name it?

@joelbqz what's the file size limit for reading and indexing, 100mb, 1gb

i'm building the fastest markdown editor, cold starts are faster than apple notes, reading and indexing files is almost as fast as the legendary sublime text, 100% free and open source launching soon

@kuxshl What was the first project where a client said, ‘Now I get it’ because of your explanation, not the design?

a lot of design content gets attention from designers, not clients. that’s the disconnect. clients are rarely impressed just because something looks clean. what they actually respond to is clarity of thought. they want to know how you see problems, how you make decisions, what your standards are, and whether there is real thinking behind the visuals. so yes, post the work, but don’t rely on the work to do everything on its own. talk about the process. talk about what was wrong before. talk about why you chose one direction over another. share opinions. share taste. share the logic. that is usually what makes the work feel valuable to someone who might actually pay for it. and then the rest is just proof. the work on the profile, the names you’ve worked with, the testimonials, the credibility. that part should be obvious when someone lands on your page. a lot of people are better than they look online simply because they are too quiet about what they’ve already done.

@vladsiu @X letss connect

Hey all - @X I'm looking to #connect with people interested in: - SaaS - Startup - Marketing Let's grow together 🤝

@vish_dev09 Usage limit's a real problem. Can't just pay to remove it either.

Claude hit usage limit. Now I have to use my brain. Unfair update. 💀

Cursor builds the UI in 2 hours. The AI layer takes 2 weeks. Not because the AI is hard. Because production AI needs architecture Cursor doesn't give you. Rate limiting, fallbacks, cost controls, hallucination guards, caching. Vibe coding skips all of it. That's the gap.

@arpit_bhayani curiosity is a powerful pull. but it can also be a time sink if it doesn't align with what you want to achieve. how do you balance that when the rabbit hole is really interesting?

Every engineer I know has asked this at some point: "How deep should I actually go?" According to me, the decision to go deep down the rabbit hole comes down to two things: 1. curiosity - what genuinely pulls you in 2. career direction - where you want to be in the next 2/3 years, not where the internet says you should be My honest take: depth works best when it serves at least one of those. Ideally, both. If something aligns with your career direction, going deep is an obvious win. One simple way to test this is to think in 2/3 year windows and ask yourself: Does understanding this layer actually move me closer to where I want to be? If you are building web apps, you do not need to master CPU instruction sets. If you are working on databases, B-tree internals matter far more than knowing every Linux kernel detail. Context changes what "deep" really means. Abstraction layers exist for a reason. They let you build without getting overwhelmed. A frontend engineer who understands HTTP is usually more valuable than one who has memorized TCP packet headers but struggles to ship features. If something does not align with your career direction, curiosity still matters. Learning out of pure interest is not wasted time. You do it because it optimizes for motivation, long-term learning, and happiness. What does not make much sense is going deep in areas that serve neither curiosity nor direction - often driven by comparison or fear. So keep checking in with yourself. Ask questions. Course-correct often. Depth is most powerful when it is intentional.

@Hi_Mrinal What's the component mapping part got you stuck on?

Pulled another all nighter to cover the research paper Novel data structures for label based queries this was genuinely most interesting read of this week, the component mapping part was a bit trickier to me tho (SKILL ISSUE) ...

@1Umairshaikh Noticed something? No. But you shipped again. Always shipping.

The loneliest part of solo building: you ship something you're proud of. nobody notices. you open cursor and start the next one.

@thesayannayak Most founders say "no competition" when they're still in discovery. They usually find out later.

We have no competition' means you didn't do any research

@TeeDevh yeah, the 'borrowing ideas' part is key. it's less about a direct lift, more like seeing how they solved a similar problem and adapting it.

Most UI that looks “wow” doesn’t come from AI. It comes from: •studying other products •borrowing ideas (not copying) •using the right frameworks & libraries •working with a good designer That’s it. #buildinpublic #saas

@siddharthwv what's the worst marketing fail you've seen

10% of selling is making something. 90% of selling is marketing it shamelessly.

@venkatofl what's wrong with radar specifically

first remove AI from stripe then give best human support for requests instead of automated templates using for support requests and and AI for payment calculation. And radar is shit.

@venkatofl what's the html edit workflow like now

Introducing Uno email editor 2.0 is live. 🔥 a faster, smarter way to build emails - with real-time previews, drag-and-drop blocks, instant HTML edits, and AI-powered suggestions. ship beautiful emails in minutes, not hours.

Free LLM tier math: Groq: 30 req/min Gemini: 15 Cerebras: 30 Mistral: 5 NVIDIA NIM: 10 Stack 3 keys each and you're at 360 req/min. For free. Enough to prototype an entire product without spending a rupee.

Gemini 2.5 Flash is a reasoning model. If you send max_tokens=1000, it burns 960 tokens on internal reasoning and gives you 40 tokens of actual output. Took me a day to figure out why my responses were getting truncated. Fix: set reasoning_effort to "none" when you don't need chain-of-thought.

@jainsahil Friends saying 'cool' is just polite noise. that’s not validation. who taught you 'no one uses it'? the data or your own bias?

Startup idea validation is brutal: You ask 3 friends. They say "cool." So you build it anyway. Then no one uses it.

Started FreeLLM as a proxy script because I kept burning through Groq's free tier every 20 minutes. Now it's at v1.5.1 with 6 providers, circuit breakers, multi-tenant keys, and response caching. Side projects that solve your own pain tend to grow faster than planned ones.

What's new in v1.5: → NVIDIA NIM support (DeepSeek R1 for free) → Tool calling streams work across all providers → Multi-tenant virtual sub-keys → Browser-safe HMAC tokens for static sites → Response cache (identical prompts return in ~23ms) → 262 tests passing Drop-in for any OpenAI SDK. Swap the base URL. Keep your code.

You shouldn't need a credit card to call an LLM. Built FreeLLM. One endpoint. 6 providers. 25+ models. Zero cost. Groq, Gemini, Mistral, Cerebras, NVIDIA NIM, Ollama. ~360 free req/min with key stacking. If one rate-limits, the next one answers. You stop seeing 429s. github.com/devansh-365/fr…