Lexpath.ai

Inspiration

As a practicing lawyer, I have seen firsthand how traditional "Legal AI" often fails the profession. Most tools are built by engineers who treat law as a search problem, but law is actually a strategy problem. I built Lexpath because I wanted a tool that didn't just find information, but actually interrogated the strength of my arguments. This is a project made by a lawyer, for lawyers—born out of the necessity to stress-test case theories in a high-stakes "War Room" environment before walking into court.

What it does

Lexpath is an Adversary Simulator designed to expose the "fragility" of a legal strategy. It uses an adversarial multi-agent system where different AI "adversaries"—a Hostile Judge, Opposing Counsel, and a Skeptical Jury—attack the user's case from every angle. It identifies logical contradictions, evidentiary gaps, and procedural weaknesses, providing a synthesized "Final Verdict" on the viability of the strategy.

How we built it

Lexpath is the first specialized component of Legant.ai, a parent project currently under construction in India. We built it using React, Vite, and Convex for a lightning-fast, real-time experience. The "brain" is powered by Gemini 2.0 Flash, choreographed with specialized models like Qwen 3 (for adversarial critique) and Saul 7B (for legal scholarship). We prioritized a "premium" aesthetic to mirror the intensity of a legal war room.

Challenges we ran into

The primary challenge was ensuring the AI adversaries maintained their distinct, hostile personas without collapsing into generic helpfulness. We also had to solve a major deployment hurdle on Vercel where custom environment variables for our Convex backend initially caused runtime crashes—a critical fix we implemented just minutes before submission.

Accomplishments that we're proud of

We are proud of creating a "living" interface that truly feels like a strategic battleground. Lexpath successfully connects a modern frontend to a production-grade backend, providing real-time adversarial reasoning that has already demonstrated its ability to find contradictions in complex legal narratives that standard LLMs miss.

What we learned

We learned that the most valuable AI for a lawyer isn't the one that agrees with them, but the one that fights them. By forcing the AI into an adversarial role, we discovered that we could extract a significantly higher "depth of reasoning" compared to standard chat interfaces.

What's next for Lexpath.ai

Lexpath is just the beginning for Legant.ai. Our roadmap involves building custom, fine-tuned LLMs for specific legal purposes, starting with the Indian legal landscape (Bharat-Legal-LLM) and eventually expanding to provide country-specific models for every jurisdiction globally. We aim to revolutionize the global legal tech stack by providing intelligence that truly understands the "spirit" and "letter" of the law in every language.

More info at : https://www.legant.tech/

Built With

• convex
• framer-motion
• github
• google-gemini-2.0-flash
• lucide-react
• pdfjs-dist
• qwen-3-(aliyun-dashscope)
• react
• saul-7b-(hugging-face)
• tailwind-css
• typescript
• vercel
• vite