AI Viva Partner：基于 Ollama Llama 3.2 的离线考试学习助手
AI Viva Partner: A Private, Offline Exam Study Assistant for My Friend
开发者用 Streamlit 和 Ollama 本地运行的 Llama 3.2 开源模型构建了 AI Viva Partner，为备考计算机考试的朋友提供离线口试模拟。该工具能生成技术题目、实时评估并打分学生答案，并指出回答不完整之处，学习笔记和成绩数据全部保留在本地。项目开源，无需调用闭源 API，因此没有 token 费用。
This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend
What I Built
I built AI Exam & Viva Partner for my friend who is preparing for computer science exams and viva rounds. It acts as an interactive, patient examiner that quizzes them on various tech topics, evaluates their answers, and provides instant feedback without needing a human study partner.
When I handed it over to my friend and ran a mock viva session, they loved how it gave instant marks and highlighted exact points where the answer was incomplete!
Demo
1. Topic Selection & Viva Question Generation
(Screenshot 2026-10-03 112120.png)
2. Live Student Answer Evaluation & Grading
(Screenshot 2026-10-03 112134.png)
3. Continuous Practice & Next Question Flow
(Screenshot 2026-10-03 112141.png)
Code
viva-partner-ai
An open-source, local AI viva & exam partner built for Hacktoberfest 2026 using Streamlit and Ollama Llama 3.2.
How I Built It
- Local Inference Engine: Used Ollama running the Llama 3.2 open-weight model locally on the machine.
- Frontend UI: Built using Streamlit for a lightweight interactive chat interface.
-
Backend Communication: Python
requestslibrary to interface with Ollama's local REST API endpoint (http://localhost:11434).
Why Does Open Innovation Matter?
- Complete Privacy: Study notes, questions, and personal performance data stay 100% offline on the user's local machine.
- Zero Cost & Infinite Usage: Running open-weight models locally eliminates expensive token costs from closed APIs.
- Custom Agent Behavior: System prompts can be modified freely to tune the examiner's strictness and behavior without API restrictions.
Prize Categories
- Overall Hacktoberfest Winner
-
Best Open-Source AI Project
来源：Google AI：DEV 作者专属（RSS） · dev.to