跳到正文
DEV Community· Mahasvan Mohan·· 2026-10-15AI 評分31

W2 Dev Challenge - MUNHelper

W2 Dev Challenge - MUNHelper

AI 導讀

MUNHelper 是為模擬聯合國代表設計的 ECOSOC 決議研究助理,透過 RAG 和開放權重 LLM 搜尋決議,並提供附有上下文引文的回答。它支援語義搜尋、更新決議文件及透過 Ollama 本機推理,讓用戶可在離線環境查找資料。作者測試指出,工具減少了翻閲 PDF 檔案的時間,並可在沒有流動網絡的模擬會議中使用。

正文

Mahasvan Mohan

Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle Submission 💸

*This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle*

What I Built

MUNHelper is a specialized preparation assistant built for Model United Nations (MUN) delegates, specifically tailored for committees under the Economic and Social Council (ECOSOC).

Model UN delegates spend dozens of hours combing through dense, historical UN resolutions, past reports, and complex policy documents to find precise clauses, legal precedents, and actionable wording for their position papers and draft resolutions. MUNHelper solves this information-overload problem by running a Retrieval-Augmented Generation (RAG) pipeline across comprehensive ECOSOC resolutions, combined with an open-weight LLM layer to answer delegate queries instantly with exact contextual citations.

Who would pay for it? Competitive high school and university MUN delegates, as well as MUN coaching academies and school clubs, who are willing to pay for a tool that cuts research time in half and gives them a competitive edge during conferences.

Demo

MunHelper

Context-based UN Resolution lookup.
Searches through every existing ECOSOC resolution and shows relevant data and extracts.

Features

  • OpenAPI-compatible API using FastAPI
  • Website using Streamlit
  • Instant ECOSOC Resolution lookup with semantic search
  • Ability to update documents with latest resolutions
  • Integration with Ollama for inference from resolution extracts
  • Docker Containerization
  • Auto Update (coming soon)

Installation with Docker

  • Install Docker Desktop and make sure Docker Compose is present.
  • Start the Docker Engine
  • Navigate to the project directory
    • cd MunHelper
  • Start the containers in detached mode using docker-compose
    •  docker-compose -f docker-compose.yml up -d
  • API is served at http://localhost:5000/docs.
  • Frontend is served at http://localhost:3000.

Usage

  • Before using anything, update the ChromaDB database
    • Visit the API Documentation at http://localhost:5000/docs.
    • Run the /manage/update-chromadb endpoint.
    • If using the Frontend
      • Hold on, I'm working on something.
    • Otherwise…

Code

You can explore the complete codebase and setup instructions directly on GitHub:

git clone https://github.com/mahasvan/munhelper
cd munhelper

How I Built It

MUNHelper is built entirely around open-source AI tooling to ensure precision, privacy, and low operational overhead:

  • Open-Weight Models: Utilizes local open-weight instruction-tuned models for text generation and summarization without depending on proprietary cloud endpoints.
  • Open-Source Orchestration & RAG: Built using open-source frameworks for document parsing, text chunking, and vector embedding generation (such as FAISS/Chroma for local vector storage and LangChain/LlamaIndex components).
  • Local Inference: The pipeline runs locally via lightweight runtime wrappers, allowing delegates to execute complex searches and context retrieval right on their laptops.

Why Does Open Innovation Matter?

Open innovation was non-negotiable for building MUNHelper for several key reasons:

  • Offline Conference Access: MUN conference venues notoriously have terrible Wi-Fi or restricted networks. Because MUNHelper runs locally with open-weight models, delegates can use it right up until committee session starts, even with zero internet connectivity.
  • Zero Per-Token API Costs: Commercial APIs charge per token, making a consumer-facing SaaS tool for students economically unviable. Running open-weight models locally keeps operating costs at zero, allowing for sustainable pricing models.
  • Privacy and Strategy Security: Delegates often input draft strategy notes and custom policy arguments alongside official documents. Keeping data processing local ensures that proprietary team strategies never leak onto third-party servers controlled by major tech companies.

Feedback

I tested MUNHelper with friends in a local college MUN team

  • What they said: They used it to cross-reference past resolutions on sustainable economic development and noted that it saved them hours of manual skimming through PDF archives. They especially liked that it worked offline during a mock committee session at a school where cell service was blocked.
  • Would they pay? They confirmed they would gladly pay a small one-time fee or seasonal subscription during conference season (September to January) to have access to a fast, reliable, offline policy research assistant.
  • What they asked for next: Exporting generated clause comparisons directly into a formatted Word or Google Doc template for draft resolutions.

Prize Categories

  • Open-Source AI Challenge Week 2: Side Hustle

來源:DEV Community · dev.to