Track: Best ERNIE Multimodal / Reasoning Application Inference Provider: Novita AI Core Models Used: ERNIE 4.5 Series (Baidu)

MANTIS is an AI-powered platform that helps developers:

• 📊 Analyze real GitHub repositories
• 🧠 Receive structured code quality feedback
• 🗺️ Generate job-ready learning roadmaps
• 🧪 Evaluate submitted code with AI mentoring
• 📈 Track learning and skill progression over time

Unlike generic AI tools, MANTIS deeply inspects actual repositories (files, README, code samples) and converts them into measurable learning intelligence.

• Downloads entire GitHub repo as ZIP

• Extracts:

  • File types & structure
  • README quality
  • Representative code samples

• Sends structured repository context to ERNIE

• Produces numerical scores + actionable insights

👉 This avoids hallucinations and ensures evidence-based evaluation

👉 Each model is chosen intentionally, not randomly.

What it does:

• Accepts any public GitHub repo

• Downloads & parses repository ZIP

• Evaluates:

  • Documentation quality
  • Code quality
  • Maintainability
  • Estimated developer level

AI Output (strict JSON):

{
  "documentation_score": 82,
  "code_quality_score": 76,
  "maintainability_score": 79,
  "estimated_developer_level": "intermediate",
  "strengths": [],
  "weaknesses": [],
  "improvement_suggestions": []
}

📌 Model used: baidu/ernie-4.5-vl-424b-a47b

Input: Any skill (e.g. Python, Computer Vision, Web Development)

Output:

• EXACTLY 10 progressive roadmap items

• Beginner → Advanced

• Each item includes:

  • Skills
  • Prerequisites
  • Hands-on projects

• Stored in DB & reusable

📌 Model used: baidu/ernie-4.5-21B-a3b-thinking

Users submit:

• Task description
• Source code

AI returns:

• Feedback list
• Improved code
• Numerical score (0–100)

Uses strict JSON contracts for reliability.

📌 Model used: baidu/ernie-4.5-21B-a3b-thinking

• Stores historical repo analysis

• Generates:

  • Progress graphs
  • Growth summaries
  • Language-wise skill estimates

This turns GitHub activity into learning analytics.

Based on repository history:

• Average scores
• Strongest languages
• Overall growth trend
• Skill maturity summary

📌 Designed for students, self-learners & early-career developers

• Python
• Flask
• SQLite
• Flask-CORS

• ERNIE 4.5 models via Novita AI
• OpenAI-compatible API
• Strict JSON-only prompting
• Regex-based safe extraction

GitHub Repo → ZIP Download
           → File & Code Parsing
           → ERNIE Analysis
           → Structured Scores
           → Learning Feedback Loop

ERINE_NOVITA/
│
├── app.py                     # Main Flask application
├── users.db                   # SQLite database
├── requirements.txt
├── runtime.txt
│
├── services/
│   ├── github_service.py      # GitHub API utilities
│   └── github_analyzer.py     # ERNIE-powered repo analysis
│
├── templates/
│   ├── index.html
│   ├── github.html
│   ├── learning.html
│   ├── profile.html
│   
│
├── static/
│   └── data/
│
└── README.md

git clone https://github.com/oneVisionary/ERINE_NOVITA.git
cd ERINE_NOVITA

python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

NOVITA_API_KEY=your_novita_api_key
FLASK_SECRET_KEY=your_secret_key

python app.py

✔ Uses real repositories, not toy examples ✔ ERNIE used as a reasoning engine, not a chatbot ✔ Strong JSON contracts & reliability handling ✔ Clear learning feedback loop ✔ Production-ready Flask architecture

• Auto PR review assistant
• Team dashboards for classrooms
• Multi-agent curriculum planning
• Resume PDF export
• Cloud deployment & CI integration

• Baidu ERNIE Team
• Novita AI
• Open-source community