ADJOURNED | Devpost

Landing page
High-Level Architecture
MCP Agent-Focused Architecture
Cute Cat

Inspiration

Lawyers specializing in financial fraud have to manage several dozens of cases at a time and spend 40% of their time on busy, non-billable work. They don't drown in law... they drown in information. Meanwhile, clients wait weeks for answers to research‑heavy questions. We built Adjourned to reduce that drag for both sides.

What it does

Adjourned is a lawyer- and client-facing workflow tool focused on financial‑fraud practice and high‑volume legal busywork.

On the lawyer-side, we summarize cases and provide recommended actions learning from human feedback, facilitate communication through AI-calls & drafted emails, and automatically complete attorney work product from client data. Adjourned also leverages an intuitive UI to allow lawyers to turn long hours into optimized case management with a 'swipe right to accept' and 'swipe left to improve' system.

On the client-side, we help people understand their rights with an AI tools that leverages RAG embeddings to eliminate hallucinations and verify accuracy. More specifically, we lean towards assisting average people fill out common forms (e.g tax and immigration) while streamlining small business productivity by teaching them tax reducing laws to make sure business is run efficiently.

Regardless of the focus, we frequently use specialized agents working in parallel to ensure quality of responses while maintaining low response times.

How we built it

Adjourned combines AI orchestration, retrieval‑augmented generation, and a responsive UI for fast human‑in‑the‑loop review. We integrated multiple services (LLMs, embeddings, and document pipelines) behind a unified workflow and support rapid iteration.

Challenges we ran into

Ensuring legal accuracy while keeping responses fast.
Designing a review flow that feels “one‑tap simple” without losing attorney control.
Keeping client explanations helpful without crossing into unauthorized legal advice.

Accomplishments that we're proud of

A working end‑to‑end flow from client intake --> drafted work product --> attorney review.
Low‑latency parallel agent system that keeps answers quick and consistent.
A UI pattern (swipe approve/improve) that makes review feel effortless.

What we learned

We learnt how important retrieval augmented generation is for ensuring legal accuracy and how to work on individual components of the project without stepping on each other's toes!