Palash Lalwani | Coding Micro-SaaS in Public

Solo dev validating tiny SaaS ideas live. First experiment: AI coach for cold emails. Sharing code, metrics, and lessons.

[Palash Lalwani | Coding Micro-SaaS in Public @PalashBuilds]

ROCm + vLLM on AMD MI300X is legitimately impressive. Qwen3-VL-32B loaded in under 5 min, 55% GPU mem util, handles multimodal inference smoothly. The 192GB VRAM headroom changes what's possible for multi-modal pipelines. @AIatAMD @lablab #AMDHackathon

[Palash Lalwani | Coding Micro-SaaS in Public @PalashBuilds]

Key insight building on @AIatAMD MI300X: run vision + audio in parallel with asyncio, deduplicate near-identical frames before the VLM, and chunk long audio at 60s. Went from 0.7x → sub 0.2x real-time factor on a 3hr Senate hearing. @lablab #ROCm #AMDHackathon

[Palash Lalwani | Coding Micro-SaaS in Public @PalashBuilds]

Building SceneIQ for @lablab AMD Hackathon: drop a 1-hour video, get a structured dossier in ~12 min. Runs Qwen3-VL-32B + VibeVoice ASR on AMD MI300X (192GB VRAM) via ROCm/vLLM. 0.2x real-time factor on long-form content 🔥 @AIatAMD #AMDHackathon

[Palash Lalwani | Coding Micro-SaaS in Public @PalashBuilds]

Just shipped OpenWhispr – a privacy-respecting dictation tool for macOS that runs entirely on your machine.

No cloud. No tracking. Pure local inference with whisper.cpp + llama.cpp.

Open source & free.

Get it here:
https://t.co/lmfw6lgZbT

[Palash Lalwani | Coding Micro-SaaS in Public @PalashBuilds]

just shipped fearmap. Type /fearmap in Claude Code and it maps your entire git repo into LOAD-BEARING / RISKY / DEAD / SAFE files

no install or api it reads your git history and tells you which files will break everything if you touch them

https://t.co/qfKRtwu7bo

@AnthropicAI

[Palash Lalwani | Coding Micro-SaaS in Public @PalashBuilds]

@SuhailKakar Cool idea but does it solve a problem? Odds are fun to watch, but why would someone open this daily vs. the actual markets or news?

[Palash Lalwani | Coding Micro-SaaS in Public @PalashBuilds]

@pika_labs Cool demo but "AI self" sounds like marketing. What’s the repeat use case after the first week? Group chats spam or something actually sticky?

[Palash Lalwani | Coding Micro-SaaS in Public @PalashBuilds]

@danielpearson @ycombinator @wideframeai Cool launch. How many minutes does it really save vs a human editor? Also curious how stable it is with brand kits, versioning, and handoffs.

[Palash Lalwani | Coding Micro-SaaS in Public @PalashBuilds]

@rork Building is the easy part. App Store rejections, device quirks, QA, and retention are the grind. How do you handle review edge cases and support when folks ship in 1 click?

[Palash Lalwani | Coding Micro-SaaS in Public @PalashBuilds]

@bricktimesyt 😂

[Palash Lalwani | Coding Micro-SaaS in Public @PalashBuilds]

Offering OpenClaw setup services.

Get running in <1 hour vs spending days on docs.

• Requirements call
• Full setup + integrations
• Documentation
• 2 weeks support

$800.

Specialize in compliance setups (legal, finance, healthcare).

DM me.

[Palash Lalwani | Coding Micro-SaaS in Public @PalashBuilds]

@bricktimesyt @albysjourney @clovr_dev @v0 @magicpatterns @Lovable Lots of people call anything they don’t understand ‘AI’. Dumb people believe everything is AI because they can’t do it or explain it. Lets see if you are smart enough to understand this

[Palash Lalwani | Coding Micro-SaaS in Public @PalashBuilds]

@dangreenheck This is the real AI productivity paradox.

You're 5X faster at execution. But now you can see 10X more opportunities. So you take on more scope.

AI doesn't reduce work. It exposes how much you left on the table because it was too hard before.

[Palash Lalwani | Coding Micro-SaaS in Public @PalashBuilds]

@AlexFinn $20K on hardware so you don't pay for tokens.

Math check: At $200/mo API costs, that's 100 months (8+ years) to break even.

Most businesses don't need 24/7 local models. Cloud API + smart usage patterns costs way less.

Privacy matters. Economics matter more.

[Palash Lalwani | Coding Micro-SaaS in Public @PalashBuilds]

@SheldonEvans Exactly.

The people larping with $20k setups will get bored in 3 months.

The people quietly shipping real products with AI will still be here in 3 years.

Hype fades. Utility compounds.

[Palash Lalwani | Coding Micro-SaaS in Public @PalashBuilds]

@nnvictory001 $3 for one task.

This is why people don't use AI agents in production. The economics don't work at scale. You need to either batch tasks or use cheaper models for routine work.

[Palash Lalwani | Coding Micro-SaaS in Public @PalashBuilds]

@aryanlabde Yes, worth it for automation. Setup takes 2-3 days DIY.

$200/month is Claude API (only if heavy use). Start with cheaper models.

I do setups for $800 if you don't want to spend the weekend on docs. DM me.

[Palash Lalwani | Coding Micro-SaaS in Public @PalashBuilds]

@anishmoonka The demo is 10% of the work.

The other 90% is all the boring stuff AI skipped: error handling, edge cases, auth, rate limits, scaling, monitoring.

AI gives you a working prototype. Production-ready is still on you.

[Palash Lalwani | Coding Micro-SaaS in Public @PalashBuilds]

@kloss_xyz This.

The demo shows potential. The real work is tuning failure modes.

Your agent will confidently do the wrong thing until you teach it not to. That's not a weekend project.

[Palash Lalwani | Coding Micro-SaaS in Public @PalashBuilds]

@LandseerEnga This is what happens when gatekeeping becomes predictable.

Apple's review guidelines are documented. The process is manual. So someone automated the check.

Every human bottleneck is a product waiting to be built.