VectoSpace · Agentic AI Education Coach

VectoSpace is a production-grade educational intelligence platform that transforms raw student data into autonomous, personalized learning journeys. Moving beyond simple grade prediction, it leverages an Agentic AI Orchestration layer to diagnose learning gaps, retrieve targeted resources via RAG, and generate interactive study plans.

---

Key Features

Feature	Description	Tech Stack
Academic Prediction	Classifies student performance (Grade 0–5) using Random Forest.	Scikit-Learn, Pandas
Status Classification	At-Risk → Below-Average → Average → Above-Average → High-Performing → Exceptional	Rule-based
Agentic Diagnosis	LLM-powered (or rule-based) learning gap analysis against student goals	LangGraph, Groq/Gemini/GPT-4o
Goal Alignment	Determines if a student is Aligned / Partially Aligned / Misaligned with their goals	Rule + LLM
Severity Scoring	Gaps rated Critical / Moderate / Minor with evidence and recommendations	Rule + LLM
Confidence Score	Data completeness score penalised for missing key fields	Rule-based
RAG Discovery	Semantic search across educational datasets to find targeted learning materials.	FAISS, Sentence-Transformers
Study Planner	Generates structured, multi-week study calendars tailored to student gaps.	LangGraph
Practice Quizzes	AI-generated interactive assessments based on identified gaps and RAG resources.	Llama-3.1, Streamlit
AI Coach Chat	Conversational interface for students to interact with their diagnosis and study plans.	Groq (Llama-3.1-8B)
Professional Exports	Automated PDF generation of comprehensive student success reports.	FPDF

---

Architecture

VectoSpace implements a sophisticated Agentic Workflow using LangGraph.

graph TD
    A[CSV Upload] --> B[ML Prediction Node]
    B --> C[Diagnosis Node]
    C --> D{LLM Available?}
    D -- Yes --> E[Agentic Gap Analysis]
    D -- No --> F[Rule-based Inference]
    E & F --> G[RAG Retrieval Node]
    G --> H[Study Planner Node]
    H --> I[Final Report Assembly]
    I --> J[User Dashboard]
    J --> K[AI Practice Quiz]
    J --> L[AI Education Coach Chat]

Loading

Core Components

• ML Engine: Pre-trained Random Forest classifier for high-accuracy grade projection.
• Inference Layer: Dual-mode (Rule-based + LLM) for robustness; falls back to deterministic rules if API keys are missing.
• Vector Store: FAISS index containing curated educational resources for RAG-driven recommendations.
• Orchestrator: LangGraph manages the state transition from raw performance data to a validated Pydantic FinalReport.

Diagnosis Engine

student_goals + performance_data
        ↓
Rule-based Engine (always runs)
        ↓
LLM (optional enhancement)
        ↓
Schema Validation
        ↓
Diagnosis Report

Prompt Strategy

agent/prompts.py implements a multi-layer prompt safety strategy:

Strategy	Implementation
Role priming	System prompt defines model as an "expert educational diagnostician"
Schema enforcement	Exact JSON structure specified with types; model outputs strict JSON
Few-shot prompting	2 labelled examples: one At-Risk, one High-Performing
Negative anchoring	Ensures model outputs [] gaps when appropriate
Guardrails	Prevent hallucination, enforce structure
Uncertainty hedging	confidence_score penalised for missing data

DiagnosisReport Schema

{
  "student_id": "STU007",
  "student_name": "Dev Patel",
  "overall_status": "At-Risk",
  "predicted_grade": "Grade 1",
  "goal_alignment": "Misaligned",
  "learning_gaps": [
    {
      "area": "Mathematics",
      "severity": "Critical",
      "evidence": "Math score is below threshold",
      "recommendations": ["...", "...", "..."]
    }
  ],
  "strengths": ["Science score is strong"],
  "priority_actions": ["Improve attendance"],
  "confidence_score": 0.9,
  "diagnosis_notes": "Multiple gaps detected",
  "source": "rule-based"
}

Project Structure

VectoSpace/
├── app.py
├── agent/
│   ├── diagnosis.py
│   ├── planner.py
│   └── prompts.py
├── src/
│   ├── agent/
│   │   ├── graph.py
│   │   ├── state.py
│   │   └── schema.py
│   └── ml/
│       ├── train.py
│       ├── retrain.py
│       ├── recommender.py
│       └── utils.py
├── rag/
│   ├── retriever.py
│   └── vectorstore_setup.py
├── extensions/
│   ├── quiz_generator.py
│   └── pdf_export.py
├── datasets/
├── model_artifacts/
├── notebooks/
├── requirements.txt
└── README.md

Setup & Installation

git clone https://github.com/crazylogic03/VectoSpace.git
cd VectoSpace

python3 -m venv venv
source venv/bin/activate

pip install -r requirements.txt

API Keys

GROQ_API_KEY="..."
GEMINI_API_KEY="..."
OPENAI_API_KEY="..."

End-to-End Walkthrough

• Ingestion: Upload dataset
• Projection: Predict grades
• Diagnosis: Identify gaps
• Intervention:
  • RAG retrieval
  • Study plan
• Assessment: Quiz generation
• Coaching: AI chat
• Certification: PDF export

Expected CSV Columns

Column	Description
student_id	Optional
student_name	Optional
attendance_percentage	%
study_hours	Weekly
math_score	Marks
science_score	Marks
english_score	Marks
internet_access	Yes/No
extra_activities	Yes/No
travel_time	Duration
parent_education	Level
gender	Category

Reliability & Guardrails

• Rule-based fallback
• Schema validation
• No data persistence
• Controlled outputs

Outputs

• Grade prediction
• Performance category
• Diagnosis
• Recommendations
• Study plan
• Quiz
• PDF report

Built for the future of personalized education.