title	Thesis Defender
emoji	🛡️
colorFrom	blue
colorTo	purple
sdk	docker
pinned	false

ThesisDefender

Paste any argument. Find out exactly where it breaks.

Built for the Microsoft Agents League Hackathon 2026 (Reasoning Agents track)

---

📖 Overview

ThesisDefender is an adversarial argument reasoning agent designed to stress-test your claims. Unlike debate tools or essay helpers, it doesn't just help you write—it dismantles what you already have and tells you exactly where your reasoning breaks, acting as a crucible for your ideas.

It takes any claim, thesis, or argument as plain text input and deploys a sophisticated, multi-step reasoning agent that:

1. Steel-Mans: Builds the strongest possible defense of the argument.
2. Attacks: Constructs the most devastating counterargument.
3. Exposes the Weakest Link: Identifies the single most fragile assumption in the argument chain.
4. Strengthens: Generates a fortified, patched version of the original claim.
5. Scores: Assigns an Argument Resilience Score (0–100) based on objective analysis.

🚀 The Constraint: 3 LLM Calls Maximum

Efficiency and precision are at the core of ThesisDefender. This project adheres to a strict limit of exactly 3 LLM API calls per analysis:

Phase	Description
Call 1 (Structure)	Extracts the core claim, sub-claims, and implicit assumptions.
Retrieval Pass	Not an LLM call. Uses Microsoft Foundry IQ to fetch real-world evidence based on the structure.
Call 2 (Dual Reasoning)	Generates both the steel-man defense and strongest attack simultaneously in one pass, utilizing the retrieved evidence.
Call 3 (Verdict)	Synthesizes the analysis into a Resilience Score, assessment, and actionable improvements.

💻 Tech Stack & Packages

We utilize a modern, highly responsive stack split into a robust asynchronous backend and a sleek frontend.

Backend (AI & API)

Powered by Python & FastAPI

• fastapi & uvicorn: High-performance async REST API framework.
• pydantic v2: For strict, robust JSON schema validation and serialization.
• redis: For async job queuing (with silent in-memory fallback for local dev).
• openai & google-generativeai: Native SDKs for model inference.
• azure-search-documents & azure-identity: Microsoft Foundry IQ integration for agentic retrieval.
• Model Support: GitHub Models (Default), OpenAI, Gemini, OpenRouter.

Frontend (UI/UX)

Powered by Next.js & React

• Next.js 14 (App Router): React framework for seamless routing and server-side rendering.
• React 18: Component-based UI logic.
• TailwindCSS: Utility-first CSS framework for a premium, responsive dark-mode aesthetic.
• Framer Motion: For fluid animations, dynamic panel transitions, and the interactive resilience meter.

🛠️ Setup & Installation

Get ThesisDefender running locally in minutes.

1. Backend Setup

1. Navigate to the backend directory:

   cd backend

2. Install dependencies:

   pip install -r requirements.txt

3. Environment Configuration: Copy .env.example to .env and configure your keys:

   MODEL_PROVIDER=openrouter
   OPENROUTER_API_KEY=your_openrouter_api_key
   # Alternatively: GITHUB_TOKEN, OPENAI_API_KEY, etc.

4. Start the Server:

   uvicorn main:app --reload

   (Optional: Start a Redis instance for durable background jobs)

2. Frontend Setup

1. Navigate to the frontend directory:

   cd frontend

2. Install dependencies:

   npm install

3. Environment Configuration: Copy .env.local.example to .env.local to point to your backend API.
4. Start the Application:

   npm run dev

🌍 Deployment

Deploy Backend (Hugging Face Spaces)

The backend is Dockerized and pre-configured for Hugging Face Spaces.

1. Create a new Docker Space on Hugging Face.
2. Link your GitHub repository or copy the contents of the backend folder.
3. In your Space's Settings, add the necessary Secrets (e.g., MODEL_PROVIDER, OPENROUTER_API_KEY).
4. The Space will automatically build and deploy on port 7860.

Deploy Frontend (Vercel)

The frontend is built with Next.js and deploys seamlessly on Vercel.

1. Go to your Vercel Dashboard and click Add New > Project.
2. Import this GitHub repository.
3. Set the Framework Preset to Next.js and the Root Directory to frontend.
4. In the Environment Variables section, add:
   • NEXT_PUBLIC_API_URL: Set this to your Hugging Face Space URL (e.g., https://your-username-spacename.hf.space).
5. Click Deploy.

🎥 Demonstration

Want to see it in action? Check out our 3-minute hackathon pitch script in DEMO_SCRIPT.md.

---

"ThesisDefender doesn't win your argument for you. It finds the hole in it — before someone else does."