Task:

Develop an AI Test Generation Bot that:

$$ Accepts multiple PDF files as input.

$$ Extracts relevant information and creates vector embeddings using Google

embeddings model and stores it in a vector database$($Pinecone$,$ Chroma etc$).$

$$ Generates objective or subjective test questions along with answers based on

user$-$specified topics, chapters, subjects, or class level through LLM$+$Semantic search

on the vector embeddings.

Specific Requirements:

PDF Processing:

$$ Implement robust PDF parsing techniques to extract text and structure accurately.

$$ Handle diverse PDF formats and layouts effectively.

Embedding Generation:

$$ Utilize any embedding model$($like Google$$s latest $001$ model$)$ understanding

capabilities to create comprehensive text embeddings.

$$ Integrate Langchain for semantic knowledge graph construction and reasoning.

$$ Incorporate Google embeddings for additional semantic context.

Test Question Generation:

$$ Develop algorithms through function calling to generate diverse and challenging test

questions aligned with specified topics and difficulty levels.

$$ Ensure questions are grammatically correct and semantically meaningful.

$$ Offer the ability to create both objective $($multiple choice, true$/$false$,$ fill$-$in$-$the$-$blank$)$

and subjective $($essay$-$style$)$ questions.

User Interface:

Design an intuitive interface for:

$$ Uploading PDF files.

$$ Specifying topics, chapters, subjects, and class levels.

$$ Generating and viewing test questions..

$$ Potentially exporting generated tests as PDF or other formats.

Evaluation Criteria:

$$ Accuracy of PDF processing and information extraction

$$ Quality of generated embeddings, capturing key concepts and relationships

$$ Relevance, clarity, and grammatical correctness of generated test questions

$$ Diversity of question types and difficulty levels

$$ Usability and intuitiveness of the user interface

$$ Ability to handle different type of content through PDF

Technologies Required:

$$ Gemini Pro or any similar language learning model

$$ Langchain framework

$$ Google embeddings

$$ Vector database$($Pinecone$,$ chroma etc$)$

$$ PDF parsing libraries $($e$.$g$.,$ PDFMiner, PyPDF$2)$

$$ User interface development framework $($e$.$g$.,$ React, Nextjs$)$

Other areas to look upon

Data quality: Emphasize the importance of high$-$quality PDF content for accurate embedding

and question generation.

Bias mitigation: Implement strategies to reduce bias in question generation, ensuring fair

and inclusive assessments.

Scalability: Consider potential architectural designs that can handle large volumes of PDF

files and generate tests efficiently.

Optional Features:

$$ Customizing test settings $($e$.$g$.,$ number of questions, difficulty level$)$

$$ chat based interface to interact with the model and generate questions as per the

prompt

Evaluation Criteria:

$$ Quality and completeness of the implemented AI Bot.

$$ Adherence to coding best practices and maintainability of the codebase.

$$ Effective storing of vast data in the form of embeddings..

$$ Accuracy and reliability of the semantic searching and quality of questions

generated.

$$ Clarity and coherence of the documentation.

$$ Attention to detail and adherence to the given requirements.