Internship Report > Workshop > AI Service (FastAPI)

AI Service (FastAPI)

AI Service Overview

Slothub’s AI Service is built using Python 3.12 + FastAPI, functioning as an automated educational content factory. The service runs on port 8000 and handles three primary tasks: exercise generation, learning roadmap creation, and document management.

Service Architecture

AI/
├── main.py                # FastAPI entrypoint - route definitions
├── services/
│   ├── openai_service.py  # OpenAI integration (embedding + generation)
│   ├── s3_service.py      # AWS S3 file upload/download
│   ├── document_service.py# PDF document processing
│   └── roadmap_service.py # Learning roadmap generation logic
├── models/                # Pydantic models for request/response
├── requirements.txt       # Python dependencies
└── .env                   # Environment configuration

Key Features

1. Automated Exercise Generation from PDFs

When a teacher uploads a PDF document, the following pipeline executes:

1. Teacher uploads PDF via Frontend
2. Frontend sends file to FastAPI endpoint
3. FastAPI stores file on AWS S3 (bucket: Slothub-upload-2026)
4. Text content is extracted from PDF
5. Content is sent to OpenAI API for analysis
6. OpenAI generates exercises/questions based on content
7. Results are returned to Frontend for display

Files are stored in a structured manner on S3 (region ap-southeast-1), using naming conventions based on classroom and subject for easy management.

2. Learning Roadmap Generation

The /roadmap endpoint analyzes a student’s current knowledge and generates a personalized study roadmap:

# Example roadmap creation flow
POST /api/v1/roadmap
{
    "subject": "Math Grade 12",
    "current_level": "intermediate",
    "target": "final exam",
    "weeks_available": 4
}

# Response: Detailed weekly roadmap
{
    "roadmap": [
        {"week": 1, "topics": ["Derivatives", "Extrema"], "exercises": 15},
        {"week": 2, "topics": ["Basic Integration"], "exercises": 20},
        ...
    ]
}

3. Chat Proxy to Slozy Agent

FastAPI acts as an intermediary proxy for chat messages, forwarding them from the frontend to the Agent-Core (Slozy AI):

Frontend → FastAPI (proxy) → Agent-Core (Slozy)
                                ↓
                          AWS Bedrock AgentCore
                                ↓
                       PostgreSQL pgvector (RAG)

OpenAI API Integration

The service uses the OpenAI API for two main purposes:

Function	Model	Description
Embedding	`text-embedding-ada-002`	Convert text to vectors for pgvector storage
Generation	`gpt-4` / `gpt-3.5-turbo`	Generate educational content (exercises, roadmaps, summaries)

# Example OpenAI call for exercise generation
import openai

response = openai.ChatCompletion.create(
    model="gpt-4",
    messages=[
        {"role": "system", "content": "You are an educational assistant..."},
        {"role": "user", "content": f"Create 5 multiple-choice questions from: {content}"}
    ]
)

AWS S3 Integration

All uploaded files are pushed to the AWS S3 bucket Slothub-upload-2026:

import boto3

s3_client = boto3.client(
    's3',
    aws_access_key_id=os.getenv('AWS_ACCESS_KEY'),
    aws_secret_access_key=os.getenv('AWS_SECRET_KEY'),
    region_name='ap-southeast-1'
)

# Upload file
s3_client.upload_fileobj(
    file_obj,
    'Slothub-upload-2026',
    f'classrooms/{classroom_id}/{filename}'
)

Direct Database Connection

Unlike the backend (which uses JPA/Hibernate), the AI service connects directly to PostgreSQL via psycopg2:

import psycopg2

conn = psycopg2.connect(
    host=os.getenv('DB_HOST'),
    dbname=os.getenv('DB_NAME'),
    user=os.getenv('DB_USERNAME'),
    password=os.getenv('DB_PASSWORD')
)

The AI service automatically runs migrations on startup — creating its own tables if they don’t exist (e.g., tables for embedding vectors). This is an intentional design allowing the service to be deployed independently.

AI Service API Endpoints

Endpoint	Method	Description
`/api/v1/generate-exercises`	POST	Generate exercises from uploaded documents
`/api/v1/roadmap`	POST	Create personalized study roadmap
`/api/v1/upload`	POST	Upload file to S3
`/api/v1/chat/proxy`	POST	Forward chat messages to Slozy
`/api/v1/subjects`	GET	List of subjects
`/api/v1/books/{id}/content`	GET	Retrieve book/curriculum content

Running the AI Service

# Navigate to the AI directory
cd AI/

# Create virtual environment
python -m venv venv
source venv/bin/activate  # Linux/Mac
# venv\Scripts\activate   # Windows

# Install dependencies
pip install -r requirements.txt

# Start the server
uvicorn main:app --host 0.0.0.0 --port 8000 --reload

The --reload flag enables hot-reload when source code changes during development. Remove this flag in production environments.

The next section explores the Agent-Core (Slozy) — the intelligence behind the AI tutor, powered by LangGraph ReAct Agent and AWS Bedrock AgentCore.