So a few days ago I shipped Klyro AI. Wrote the launch post, got the LinkedIn validation hit, moved on. And I mean moved on — like, the same week, because that's just how my project list works. 50+ projects before I'm done with college isn't a number I get to be precious about. It only means something if I keep shipping.
But this next one wasn't planned the way Klyro was. It started kind of annoying, actually.
I was trying to settle some random argument — can't even remember what, something about a President's ordinance power — and I went down the usual rabbit hole. Three Quora answers that contradicted each other. A law school PDF from probably 2014. A "legal explainer" site that was 80% ads and 20% nothing. And the whole time I'm thinking: the actual answer is sitting in the Constitution itself. 395 articles, 22 parts, public document, free to read. Nobody's hiding this from me.
Except nobody reads a 395-article document to settle one argument. That's the actual gap. Not "people don't know their rights" in some abstract sense — it's that the source of truth exists and is completely unusable in practice.
That annoyance turned into LexByte AI Lawyer.
🔗 Live: https://lexbyte-ai-lawyer.web.app/
What it does
Two things, and I kept it to two on purpose.
First, you can just ask it stuff. Plain language, no legal Latin required. "Explain Article 21 like I'm not a law student." "What's the basic structure doctrine." "Fundamental Rights vs Directive Principles, what's actually different." It answers in a structured way and points back to the specific Articles it's pulling from — it's not supposed to just sound confident, it's supposed to show its work.
Second, there's a full searchable database of the Constitution itself. By Part — Fundamental Rights, Directive Principles, Emergency Provisions, all of it. Search by article number, title, or keyword. Sometimes you don't want a conversation, you just want to go straight to Article 21 and read it yourself. Both modes exist because both are real use cases.
I also spent more time than I'd like to admit on the UI being calm. Glassmorphism, soft, almost like a counsel's office rather than a search engine. Legal stuff is already stressful for most people — the interface's whole job is to not add to that.
Why I picked something this specific
I could've made a "general AI lawyer for everything everywhere," and it would've been worse on day one than LexByte is right now. Here's the thing nobody tells you when you start these projects: a narrow tool that's actually right beats a broad tool that's occasionally right, every single time, especially in law.
The Constitution is a closed corpus. Finite. 395 articles. That's a problem I can actually solve well instead of gesturing at solving. And honestly — if a legal AI tool makes up an article number, that's not a "minor inaccuracy," that's the entire product failing at the one job it had. So getting answers grounded in real text, not just plausible-sounding text, was the actual hard part of this build. Not the chat UI. The trust part.
Where it actually is right now, no sugarcoating
It's a static site right now — HTML, CSS, vanilla JS, Firebase Hosting. No backend spinning, no install, just a link. I'm not training or fine-tuning any model here — that was never the plan. I'm using an LLM API for the actual language understanding and reasoning, and feeding it structured knowledge of the Constitution as context, so it's answering from the real article text instead of whatever it half-remembers from training data. Currently the corpus has 179+ articles indexed and I'm working toward the full 395. That's the part that makes the citations real instead of decorative — the model isn't guessing what Article 21 says, it's being handed the actual text.
It's intentionally simple right now. I'd rather launch the smaller, correct version and grow it in public than launch something that sounds impressive and falls apart on the third question. The roadmap's where the ambition actually lives:
- Case law — Supreme Court and High Court judgments, not just the Constitution
- Real citation generation
- Multi-language support, because rights shouldn't require fluency in English
- AI contract review and document analysis
- A proper personalized research workspace
- PDF export for whatever you're working on
Who I actually built this for
Law students cross-referencing provisions at 1am before an exam. Paralegals doing the boring-but-necessary prelim research. Teachers who want students to actually engage with constitutional structure instead of zoning out at a textbook. And just regular people — most of us genuinely don't know what our Fundamental Rights say until we're standing somewhere needing them.
One thing I'm not going to soften: this is not a lawyer. The disclaimer stays visible in every session on purpose — this is legal information, not legal advice, and that line doesn't move for anyone.
What's next
LexByte's MIT-licensed, so if you want to dig into the article corpus, the search layer, or eventually case law integration, the door's open. This is project 13 out of 23 on my roadmap — next up is the heavier system-design stuff, a mini browser, a full browser engine, a mini IDE. I'll keep posting as things ship, same as always.
If you've actually built grounded AI retrieval over structured legal or regulatory text before, I want to hear how you approached it — drop it in the comments, I'm not above stealing good ideas.
Try it: https://lexbyte-ai-lawyer.web.app/
