Business Problem

In advanced education, students preparing for highly competitive exams like JEE Advanced, NEET, SAT and Gaokao face exceptionally complex questions. Traditional methods of solving these doubts are often inefficient and time-consuming. There is a significant need for a detailed, explainable doubt solver that can provide accurate and understandable solutions to these intricate problems and also students can follow up at any step which helps students grasp complex concepts effectively.

About Doubt Buddy

Doubt Buddy is one of India’s leading education companies for students of classes 6 to 12 and engineering and medical entrance exam aspirants, Doubt Buddy provides students with personalized practice, mock tests, previous year question papers, and live doubt resolution which help students learn better and faster. Doubt Buddy covers the syllabus for all school boards across India for Class 6 to 12.

Solution

To address the challenge, we developed a sophisticated chatbot system with the following components:

Doubt Buddy, a groundbreaking project funded by the Ministry of Education, is revolutionizing how students tackle complex academic problems. At the heart of Doubt Buddy's innovation is an advanced AI system designed to transform the learning experience for students preparing for some of the world's toughest exams.

Leveraging cutting-edge AI technologies, Doubt Buddy has developed a sophisticated doubt-solving system centered around a state-of-the-art decision tree-based Llama-3 8B fine-tuned model. This isn't just any model it's meticulously trained on a colossal dataset of 3 million complex problems and solutions, curated by expert teachers from around the globe. This extensive training ensures that the model doesn't just memorize solutions but understands and applies expert-level problem-solving strategies.

Doubt Buddy's approach goes beyond traditional fine-tuning. By incorporating advanced decision tree-based methodologies, the model can simulate the critical thinking process of a seasoned educator. This enables it to break down complex problems into manageable steps, providing students with detailed, step-by-step, accurate, and easy-to-understand solutions.

Architecture

• Users Interaction
  - Users access the system through a web interface.
  - Requests are directed to the AWS infrastructure.
• Web Application Firewall (WAF)
  - Protects against common web exploits and attacks.
  - Ensures only legitimate traffic reaches the web server.
• Public Subnet
  - Hosts the Web server that handles incoming user requests.
  - Web server communicates with the API Gateway to forward requests.
• API Gateway
  - Serves as the main entry point for all user requests.
  - Routes requests to the Lambda function for processing.
• Lambda Inference
  - Processes requests using FastAPI and LyzrSDK.
  - Executes serverless functions to handle inference tasks.
• Private Subnet
  - Amazon S3: Stores documents and static assets.
  - EC2 Instances: Perform intensive data processing and computations.
• Data Processing and Machine Learning
  - Processing and Embedding:
    - Processes documents from S3 to create embeddings.
    - Uses EC2 instances to handle data transformation.
  - Finetuning:
    - Fine-tunes ML models (e.g., LLAMA2) using embeddings.
    - Stores processed results in Postgres for persistent storage.
• Version Control and Deployment
  - Git: Manages the source code and configuration files.
  - ECR: Hosts containerized applications for deployment.
• Monitoring and Management
  - CloudWatch: Monitors system performance, logs, and metrics.
  - Secrets Manager: Manages and secures sensitive information like API keys and credentials.