Skip to content

iihatem/knighthacks2025

BranchesTags

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

29 Commits
.github
.github
 
 
backend
backend
 
 
frontend
frontend
 
 
.gitignore
.gitignore
 
 
README.md
README.md
 
 

Repository files navigation

Aura MM — AI-Powered Legal Case Management

An intelligent legal case management platform built for Morgan & Morgan, featuring multi-agent AI orchestration, document RAG search, and human-in-the-loop approval workflows. Built at Knight Hacks 2025.

Overview

Aura MM assists lawyers and paralegals with end-to-end case management using a fleet of specialized AI agents. From processing evidence and drafting client emails to conducting legal research and scheduling depositions — every sensitive action is routed through an approval workflow before execution.

Features

  • Case Management — Create and manage cases with a MM-YYYY-NNNNN numbering scheme
  • Document Ingestion & RAG — Ingest PDFs and images into a Snowflake vector store; semantic search across all case documents
  • Multi-Agent Orchestration — An orchestrator routes user queries to specialized sub-agents
    • Legal Researcher — Gemini 2.5 Pro with Google Search grounding; finds precedents, citations, and rates case strength
    • Client Communication Guru — Drafts empathetic client emails; requires approval before sending
    • Evidence Sorter — Processes email attachments and categorizes evidence
    • Records Wrangler — Searches records via RAG and drafts record request emails
    • Voice Scheduler — Schedules depositions and mediations via Google ADK
  • Activity Approval Workflow — All sensitive agent actions (emails, scheduling) are logged as pending activities requiring human approval before execution
  • Session Management — Persistent, case-scoped conversation history across all agents
  • Salesforce Integration — Case and task management synced with Salesforce CRM

Tech Stack

Layer Technology
Frontend Next.js 16, React 19, TypeScript, Tailwind CSS v4
Backend Python, Flask
Database Snowflake (vector embeddings via EMBED_TEXT_768)
AI / LLM Google Gemini 2.5 Flash & Pro, Gemini Vision
File Storage DigitalOcean Spaces (S3-compatible)
CRM Salesforce
Email SMTP (Gmail)
Voice ElevenLabs
Scheduling Google ADK (Agents Development Kit) + Google Calendar API

Project Structure

knighthacks2025/
├── backend/
│   ├── app.py                        # Main Flask application & all API routes
│   ├── requirements.txt
│   ├── services/
│   │   ├── activity_logger.py        # Logs agent actions to Snowflake
│   │   ├── session_manager.py        # Conversation session management
│   │   ├── email_sender.py           # SMTP email service
│   │   ├── evidence_processor.py     # Email attachment processing
│   │   └── salesforce_service.py     # Salesforce API integration
│   ├── agents/
│   │   └── orchistrator_agent/
│   │       ├── agent.py              # Routes queries to sub-agents
│   │       └── sub_agents/
│   │           ├── legal_researcher_agent/
│   │           ├── client_communication_agent/
│   │           ├── evidence_sorter_agent/
│   │           ├── records_wrangler_agent/
│   │           └── voice_scheduler_agent/
│   └── tools/
│       ├── communication_tools.py
│       └── salesforce_tools.py
└── frontend/
    ├── src/
    │   ├── app/                      # Next.js App Router pages
    │   │   ├── page.tsx              # Dashboard
    │   │   ├── cases/
    │   │   ├── case/[slug]/
    │   │   ├── clients/
    │   │   ├── client/[slug]/
    │   │   ├── calendar/
    │   │   ├── analytics/
    │   │   └── settings/
    │   ├── components/
    │   │   ├── Dashboard.tsx
    │   │   ├── GlobalChat.tsx        # Chat interface for agents
    │   │   ├── AddCaseModal.tsx
    │   │   └── ApprovalCard.tsx
    │   ├── lib/api.ts                # API client
    │   ├── hooks/
    │   │   ├── useCases.ts
    │   │   └── useRAG.ts
    │   └── types/case.ts
    └── package.json

Getting Started

Prerequisites

  • Python 3.12+
  • Node.js 18+
  • A Snowflake account with vector search enabled
  • Google Gemini API key
  • DigitalOcean Spaces bucket (or any S3-compatible storage)

Backend

cd backend
python -m venv venv
source venv/bin/activate        # Windows: venv\Scripts\activate
pip install -r requirements.txt
cp .env.example .env            # Fill in your credentials
python app.py                   # Starts on http://localhost:5001

Frontend

cd frontend
npm install
npm run dev                     # Starts on http://localhost:3001

Environment Variables

Create a backend/.env file with the following:

# Google Gemini
GEMINI_API_KEY=
GEMINI_MODEL=gemini-2.5-flash

# Snowflake
SNOWFLAKE_ACCOUNT=
SNOWFLAKE_USER=
SNOWFLAKE_PASSWORD=
SNOWFLAKE_DATABASE=
SNOWFLAKE_SCHEMA=
SNOWFLAKE_WAREHOUSE=
SNOWFLAKE_ROLE=

# DigitalOcean Spaces (S3-compatible)
DO_SPACES_KEY=
DO_SPACES_SECRET=
DO_SPACES_BUCKET=
DO_SPACES_REGION=

# SMTP Email
FIRM_EMAIL=
FIRM_EMAIL_PASSWORD=
SMTP_SERVER=smtp.gmail.com
SMTP_PORT=587
FIRM_NAME=

# Salesforce
SALESFORCE_CLIENT_ID=
SALESFORCE_CLIENT_SECRET=
SALESFORCE_INSTANCE_URL=
SALESFORCE_USERNAME=
SALESFORCE_PASSWORD=
SALESFORCE_SECURITY_TOKEN=

# ElevenLabs (voice)
ELEVENLABS_API_KEY=

API Reference

Method Endpoint Description
GET / Health check
POST /api/create-case Create a new case
GET /api/list-cases List all cases
GET /api/view-case-data/<case_id> Get full case details and document chunks
POST /api/add-case-files Upload additional files to a case
POST /api/test-rag Semantic search across case documents
POST /api/agent/process Send a message to the orchestrator agent
GET /api/activities/<case_id> Get agent activities for a case
POST /api/activities/<activity_id>/approve Approve and execute a pending activity
POST /api/activities/<activity_id>/reject Reject a pending activity
GET /api/activities/pending Get all pending approvals
GET /api/sessions/<case_id> Get sessions for a case
GET /api/sessions/<session_id>/messages Get conversation history
POST /api/sessions/<session_id>/end End a session
POST /api/evidence/process-email Process an incoming email for evidence

Agent Architecture

User Message
      │
      ▼
[Orchestrator Agent]
      │  Analyzes intent + session history + case context
      │
      ├──► Legal Researcher       (Gemini 2.5 Pro + Google Search grounding)
      ├──► Client Communication   (email drafting → approval → SMTP send)
      ├──► Evidence Sorter        (email attachment categorization)
      ├──► Records Wrangler       (RAG search + record request emails)
      └──► Voice Scheduler        (Google ADK + Calendar API)
                │
                ▼
        [Activity Logger]         (if action requires human approval)
                │
                ▼
        Pending Approval Queue ◄── Lawyer approves/rejects in UI
                │
                ▼
           Execute Action

Sensitive actions (sending emails, scheduling calls, creating Salesforce tasks) are never executed automatically — they are logged as pending activities that must be approved by a lawyer before execution.

Database Schema (Snowflake)

Table Purpose
case_data Case metadata, document text chunks, and 768-dim vector embeddings
agent_activities All agent actions with pending/approved/rejected/completed status
agent_sessions Conversation sessions scoped to a case
agent_messages Full message history within sessions

Vector similarity search uses Snowflake's native VECTOR_COSINE_SIMILARITY and EMBED_TEXT_768 functions.

Frontend Pages

Route Page
/ Dashboard — recent cases, pending approvals, quick actions
/cases All cases with filtering
/case/[id] Case detail — documents, chat, activities
/clients Client management
/client/[id] Individual client profile
/calendar Schedule depositions and meetings
/analytics Case metrics and analytics
/settings Application configuration

Built With

About

No description, website, or topics provided.

Resources

Readme
Activity

Stars

0 stars

Watchers

1 watching

Forks

1 fork
Report repository

Releases

No releases published

Packages

 
 
 

Contributors

Languages