Case Closed

As part of the AI Hackathon, we built Case Closed — an AI-powered precedent recommendation engine that blends multi-agent reasoning, modern LLMs, and the CourtListener API to make legal research faster and more contextual. The project aims to move beyond keyword search and toward legal-similarity-driven retrieval, automated memo/brief drafting, and an iterative, chat-style research workflow.

Contributors:

• Sai Yadavalli — AI Engineer
• Sedat Unal — Full-stack Developer
• Jason Pereira — Frontend Developer & UI/UX Designer
• Saksham Anand — Backend Developer

---

Inspiration — Make legal research feel like conversation

Legal research often requires sifting through thousands of opinions to find a single meaningful precedent. We wanted to reduce that friction. Instead of treating legal documents as strings to be keyword-matched, Case Closed reasons about facts, issues, and legal posture — more like a junior associate guided by a senior. Our goal was to create a pipeline that could:

• extract structured legal elements from uploads or prompts,
• query live case databases intelligently, and
• produce human-readable explanations and legal drafts.

The result is a research assistant that helps you find relevant cases, explains why they matter, and helps convert findings into memos and briefs — all within a single session.

---

Pitch

Case Closed is a conversational legal research assistant. Users upload a court opinion (PDF / text) or describe their fact pattern; the system extracts facts, identifies legal issues and jurisdiction, constructs targeted queries to CourtListener, ranks results by contextual similarity (not just lexical overlap), and produces explainers and draft legal memos.

This approach accelerates legal workflows by combining: multi-agent reasoning (fact extraction, query generation, relevance scoring, drafting), semantic retrieval (embeddings + Gemini), and direct integration with CourtListener for authoritative case text.

---

Overview

The repository contains the full-stack app, agent pipeline, and demo frontend.

• app.py — Flask server and agent orchestrator
• app.py/ — modular agent definitions (extractor, retriever, evaluator, drafter)
• static/ — vanilla JavaScript/CSS chat-style UI and upload interface
• templates/ — main HTML files

Features

• Contextual Precedent Retrieval: Retrieves cases from CourtListener using structured queries derived from extracted facts and issues.
• Multi-Agent Pipeline: Separate agents for extraction, search-query construction, retrieval, evaluation, and drafting.
• Draft Generation: Generate memos and briefs based on retrieved cases and session context.
• Session Persistence: Each user session stores context, enabling iterative refinement and progressive drafting.
• Filterable Results: Narrow results by jurisdiction, court level, and issue similarity.
• Easy Uploads: Upload PDFs or paste text — the system extracts and analyzes content automatically.

---

Example walk-through

1. Upload a case PDF or paste a fact pattern.
2. The Extraction Agent pulls parties, claims, dates, and core issues.
3. The Query Agent turns those structured facts into optimized CourtListener queries.
4. The Retrieval Agent fetches candidate opinions and embeddings are used to compute semantic similarity.
5. The Evaluation Agent ranks and explains why each result is relevant.
6. Ask the Drafting Agent to produce a memo or brief section; iterate until satisfied.

---

Stack Overview

---

Demo

Click the video below to view our skit and demo!

Setup and Running

Prerequisites

1. Install Docker Desktop

   • Download and install Docker Desktop
   • Ensure Docker is running before proceeding

2. Required Files

   • .env file with your environment variables (PROJECT_ID, GOOGLE_CLOUD_LOCATION, COURTLISTENER_TOKEN, GOOGLE_APPLICATION_CREDENTIALS=/key.json)
   • key.json - Google Cloud service account credentials file in the project root directory. Paste your key in the file once you acquire it.

Running on Mac/Linux

1. Build the Docker image:

   docker build --no-cache -t caseclosed .

2. Run the container:

   docker run -p 5000:5000 --env-file .env -v "$(pwd)/key.json:/app/key.json" -e GOOGLE_APPLICATION_CREDENTIALS=/app/key.json caseclosed

3. Access the application:

   • Open your browser and navigate to: http://localhost:5000

Running on Windows

1. Clone This Repository

2. Build the Docker image:

   docker build --no-cache -t caseclosed .

3. Run the container (PowerShell):

   docker run -p 5000:5000 `
     --env-file .env `
     -v "${PWD}/key.json:/app/key.json" `
     -e GOOGLE_APPLICATION_CREDENTIALS=/app/key.json `
     caseclosed

   Or using Command Prompt:

   docker run -p 5000:5000 --env-file .env -v "%cd%/key.json:/app/key.json" -e GOOGLE_APPLICATION_CREDENTIALS=/app/key.json caseclosed

4. Access the application:

   • Open your browser and navigate to: http://localhost:5000

Stopping the Container

Press Ctrl+C in the terminal where the container is running, or:

docker ps  # Find the container ID
docker stop <container-id>