MartusAI

Inspiration

Millions of Americans receive legal documents they can't understand — eviction notices, wage letters, court summons. Most never respond because they don't know what the document means. Legal aid is overwhelmed and lawyers cost $300+/hour. We built MartusAI to close that gap.

What it does

MartusAI is capable of instantly converting any legal document into plain language answers, citing sources and providing a ready-to-reply response letter within 30 seconds. Just upload a PDF or image and get an easy-to-understand summary, all claims cited to statutes, a deadline timeline, and a response letter ready to send. Four modules – PlainSite, ContractFlow, EvidenceChain and TenantShield - translate, analyze, organize and protect your rights through legal documents.

How we built it

Built with Next.js 14, React, TypeScript, Tailwind, and shadcn/ui on the frontend, and Supabase (PostgreSQL) for storage — xAI Grok is the engine for computer vision/text analysis of the pipeline, and Groq (Llama 3.3 70B) for the four modules. The OCR is managed via Tesseract.js and PDF extraction with pdf.js, and is deployed on Vercel at martusai.xyz.

Challenges we ran into

DNS migration was brutal — moving from Cloudflare to Vercel to Netlify and back, each with different SSL and proxy requirements. We also had to architect a clean split between two LLM providers (Grok for vision, Groq for text) and build a citation-grounding layer so no claim is ever hallucinated.

Accomplishments that we're proud of

Shipped a fully working prototype in under 36 hours. It's live at martusai.xyz, with a real domain. All legal claims are cited, privacy focused, has an anonymous mode and is mobile first for people reading on their phone with a deadline looming.

What we learned

The Legal Tech problem is more design than engineering, as much as we'd like perfect AI to handle citations. Perfect AI is not going to happen. You can have two LLMs power the same system without any issues so long as there's sufficient abstraction between them. DNS is the real pain in the hackathon project.

What's next for MartusAI

The first major update will expand from 10 thoroughly-researched states to all 50, while also introducing a native mobile application with offline functionality, localization in Spanish and additional languages, and partnerships with legal aid clinics, libraries and tenant unions. Looking further ahead, we're working on courtroom preparation mode and voice input options for the visually impaired.

Built With

• grok
• next.js
• react
• supabase
• typescript
• vercel