██████╗  █████╗ ████████╗████████╗██╗     ███████╗     █████╗ ██████╗ ███████╗███╗   ██╗ █████╗ 
  ██╔══██╗██╔══██╗╚══██╔══╝╚══██╔══╝██║     ██╔════╝    ██╔══██╗██╔══██╗██╔════╝████╗  ██║██╔══██╗
  ██████╔╝███████║   ██║      ██║   ██║     █████╗      ███████║██████╔╝█████╗  ██╔██╗ ██║███████║
  ██╔══██╗██╔══██║   ██║      ██║   ██║     ██╔══╝      ██╔══██║██╔══██╗██╔══╝  ██║╚██╗██║██╔══██║
  ██████╔╝██║  ██║   ██║      ██║   ███████╗███████╗    ██║  ██║██║  ██║███████╗██║ ╚████║██║  ██║
  ╚═════╝ ╚═╝  ╚═╝   ╚═╝      ╚═╝   ╚══════╝╚══════╝    ╚═╝  ╚═╝╚═╝  ╚═╝╚══════╝╚═╝  ╚═══╝╚═╝  ╚═╝

The High Court of Artificial Intelligence — where AI models argue, and an AI judge decides.

---

⚖️ Order in the Court

AI Battle Arena is a courtroom-themed AI battle platform. You submit a prompt — called a Brief — and two AI models argue their case side by side. A third AI model sits as the Presiding Judge and delivers a binding verdict with scores and judicial reasoning.

No human bias. No subjective opinion. Pure AI vs AI — judged by AI.

🏛️ The Courtroom — UI Breakdown

Every element of the interface is designed around a real courtroom. This isn't just a theme — it's the entire interaction model.

Element	Role	Description
⚖️ The Bench	Presiding Judge	Google Gemini AI — delivers the final verdict with score and judicial reasoning
🏛️ Witness Box I & II	Counsel AI-1 & AI-2	Mistral AI (Defense) and Cohere AI (Prosecution) present their answers side by side
📁 Case File	Center Panel	Displays the litigation subject (your prompt) — the matter before the court
🎙️ The Podium	Input Bar	Where you submit your brief — the bottom input that initiates the session
📋 Court Ledger	History Drawer	Collapsible panel (top right) that stores all past dockets and sessions
📜 Dockets	Prompt History	Previous prompts listed under the Case File — select any to revisit that battle
🔴 Judicial Finding	Verdict Card	Score badge + judge's written reasoning rendered under each Witness Box

🧠 The Intelligence Stack

Under the hood, the courtroom runs on a stateful multi-agent LangGraph pipeline — a graph-based execution engine that coordinates three AI agents with full control over parallel dispatch, state, and verdict aggregation.

LangChain acts as the universal adapter layer so every model receives the exact same prompt under identical conditions — zero bias, maximum fairness.

                        ┌──────────────────────────────┐
                        │    Brief (User Prompt)        │
                        │    Submitted at The Podium    │
                        └──────────────┬───────────────┘
                                       │
                        ┌──────────────▼───────────────┐
                        │     LangGraph Orchestrator    │
                        │     (Stateful Agent Graph)    │
                        └───┬──────────────────────┬───┘
                            │  parallel dispatch   │
              ┌─────────────▼──┐              ┌────▼──────────────┐
              │  Witness Box I  │              │  Witness Box II    │
              │  Mistral AI     │              │  Cohere AI         │
              │  (Defense)      │              │  (Prosecution)     │
              └─────────────┬──┘              └────┬──────────────┘
                            │                      │
                        ┌───▼──────────────────────▼───┐
                        │         THE BENCH             │
                        │      Google Gemini AI         │
                        │      (Presiding Judge)        │
                        │  Scores + Judicial Reasoning  │
                        └───────────────────────────────┘
                                       │
                        ┌──────────────▼───────────────┐
                        │    Verdict Delivered           │
                        │    Logged to Court Ledger      │
                        └──────────────────────────────┘

🛠️ Tech Stack

Backend Runtime — Node.js + TypeScript (ESM) Framework — Express v5 AI Orchestration — LangGraph (stateful multi-agent graph) AI Framework — LangChain (unified provider interface) AI Providers — Google Gemini · MistralAI · Cohere Schema Validation — Zod Database — MongoDB via Mongoose Logging — Morgan Deploy — Render

Frontend Framework — React 19 Build Tool — Vite 7 Styling — Tailwind CSS v4 Scroll — Locomotive Scroll (smooth scroll) Theme — Vintage Courtroom (custom hand-crafted) Architecture — Single Page Application Deploy — Render

🔬 How LangGraph Runs the Trial

Each battle is a compiled LangGraph graph with discrete, stateful nodes:

• Dispatch Node — Fans out the brief to both Counsel nodes simultaneously in parallel
• Counsel Nodes — Mistral and Cohere run as isolated agents with independent retry and timeout handling
• Judge Node — Google Gemini receives both responses and evaluates them, returning a score and written judicial finding for each
• Aggregation Node — Collects the full trial record, normalizes it via LangChain's unified schema, and commits it to the Court Ledger
• State — The entire session (prompt → responses → verdict → scores) flows through a typed state object, making every trial reproducible

Adding a new AI model = adding a new Counsel node. The judge and orchestration logic stay untouched.

🤖 Court Constituents

Role	Model	Provider
⚖️ Presiding Judge	Gemini	Google
🛡️ Defense Counsel	Mistral	MistralAI
⚔️ Prosecution Counsel	Command R	Cohere

---

"UPON DUE CONSIDERATION of the compelling pleadings filed herein..."

The Court is now in session.

Made by Notanormaldev