EduMentor-Qwen3-4B-v2-FP16/README.md

---
license: apache-2.0
language:
- en
pipeline_tag: text-generation
library_name: transformers
base_model: PraneetNS/EduMentor-Qwen3-4B-FP16
tags:
- qwen3
- education
- engineering
- mentor
- llm
- conversational
- multimodal
- speech
- json
- fine-tuned
---

# EduMentor Qwen3 4B v2 (FP16)

EduMentor v2 is the second major release of EduMentor, an AI engineering mentor built for university students.

The model is designed to provide conversational technical guidance, placement preparation, project mentoring, and structured educational responses across multiple engineering disciplines.

Unlike a generic chatbot, EduMentor is optimized for voice-first tutoring systems where spoken explanations are separated from visual artifacts such as code, diagrams, roadmaps, tables, and notes.

---

# Highlights

## New in Version 2

Compared to EduMentor v1, this release includes:

- Fine-tuned on an expanded multi-turn engineering conversation dataset (~27K conversations).
- Improved contextual follow-up handling.
- Better conversational flow for tutoring sessions.
- Stronger identity consistency as EduMentor.
- Improved reasoning across Computer Science and core engineering subjects.
- Enhanced placement and career guidance.
- Better structured JSON responses for multimodal applications.
- Verified merged FP16 checkpoint (no LoRA dependency).

---

# Model Information

| Property | Value |
|----------|-------|
| Base Model | EduMentor-Qwen3-4B-FP16 (v1) |
| Architecture | Qwen3-4B |
| Fine-tuning | Supervised Fine-Tuning (LoRA) |
| Merge | Fully merged FP16 |
| Context Length | 4096 tokens |
| Precision | FP16 |
| Intended Use | Engineering Mentor |

---

# Training Dataset

EduMentor v2 was trained on approximately 27,000 carefully curated multi-turn conversations covering engineering education.

The dataset emphasizes:

- realistic mentor-student interactions
- conceptual teaching
- problem solving
- project guidance
- interview preparation
- career mentoring
- emotional encouragement
- structured responses

The conversations include contextual follow-up questions to simulate natural tutoring sessions.

---

# Supported Domains

## Computer Science

- Programming Fundamentals
- Object Oriented Programming
- Data Structures
- Algorithms
- Operating Systems
- DBMS
- Computer Networks
- Software Engineering

---

## Artificial Intelligence

- Machine Learning
- Deep Learning
- Neural Networks
- LLMs
- Transformers
- RAG
- Prompt Engineering
- AI Deployment

---

## Electronics

- Digital Electronics
- Analog Electronics
- Signals
- Communication
- Embedded Systems
- Microprocessors

---

## Electrical Engineering

- Machines
- Power Systems
- Control Systems
- Power Electronics

---

## Mechanical Engineering

- Thermodynamics
- Manufacturing
- Design
- Strength of Materials
- Fluid Mechanics

---

## Civil Engineering

- RCC
- Structural Engineering
- Surveying
- Transportation
- Environmental Engineering

---

## Mathematics

- Calculus
- Linear Algebra
- Probability
- Statistics
- Discrete Mathematics

---

## Career Guidance

- Placements
- Resume Reviews
- Internship Guidance
- Interview Preparation
- Learning Roadmaps
- Project Ideas

---

# Response Format

EduMentor is designed for multimodal tutoring systems.

Typical responses follow the format:

```json
{
  "speech": "...",
  "display": {
    "type": "code | notes | roadmap | table | flowchart",
    "content": "..."
  },
  "follow_up": "..."
}
```

This allows downstream applications to:

- speak only the explanation
- display diagrams separately
- avoid reading code aloud
- render structured educational artifacts

---

# Intended Voice Pipeline

```
User Speech
      │
      ▼
Speech Recognition
      │
      ▼
EduMentor v2
      │
      ▼
JSON Parser
      │
 ┌────┴─────────────┐
 ▼                  ▼
Speech           Display
(TTS)         (Code / Notes / Roadmaps)
```

Recommended stack:

- Faster-Whisper
- EduMentor
- llama.cpp
- Kokoro TTS

---

# Example

## User

Explain Binary Search.

## Assistant

```json
{
  "speech": "Binary Search repeatedly divides the search space in half, making it much faster than linear search on sorted data.",

  "display": {
    "type": "code",
    "language": "python",
    "content": "def binary_search(...): ..."
  },

  "follow_up": "Would you like to understand the time complexity?"
}
```

---

# Usage

```python
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model_id = "PraneetNS/EduMentor-Qwen3-4B-v2-FP16"

tokenizer = AutoTokenizer.from_pretrained(model_id)

model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.float16,
    device_map="auto"
)
```

---

# Intended Applications

EduMentor is suitable for:

- AI Tutors
- Educational Chatbots
- Voice Assistants
- Engineering Learning Platforms
- Placement Preparation
- Career Mentoring
- Project Guidance
- Classroom Assistants

---

# Limitations

EduMentor may occasionally:

- generate incorrect technical information
- require verification for safety-critical engineering tasks
- produce imperfect JSON formatting for highly complex requests
- benefit from external tools or retrieval for rapidly changing topics

It should not be considered a replacement for certified professional engineering advice.

---

# Roadmap

Future versions aim to include:

- Tool Calling
- Retrieval-Augmented Generation (RAG)
- Long-Term Student Memory
- Personalized Learning Plans
- Multimodal Diagram Generation
- Real-Time Coding Assistance
- Agentic Workflows

---

# Citation

If you use EduMentor in academic work or projects, please cite this repository.

---

# Acknowledgements

EduMentor is built upon the Qwen3 architecture and fine-tuned to provide personalized engineering education through conversational AI.

---

# Creator

**Praneet N S**

EduMentor is an ongoing effort to build an AI mentor capable of assisting engineering students through natural conversations, structured explanations, and voice-first educational experiences.