Skip to content

README.md

Latest commit

 

History

History
96 lines (73 loc) · 3.72 KB

README.md

File metadata and controls

96 lines (73 loc) · 3.72 KB

LitAI: Structural Manuscript Auditor

LitAI is an industry-grade, AI-powered literary assistant designed to perform deep structural audits on manuscripts. Unlike basic summarization tools, LitAI utilizes an Agentic RAG (Retrieval-Augmented Generation) pipeline to simulate the mind of a Senior Developmental Editor, identifying narrative inconsistencies and thematic gaps with precise source attribution.

🚀 Key Features

  • Semantic Section Marking: Automatically identifies and cites specific page numbers (e.g., [📍 Page 12]) using metadata-aware retrieval.
  • Editor Persona Engine: Leverages Gemini 3 Flash with thinking_level="medium" to generate critical questions, structural impacts, and expected narrative answers.
  • Fluid Intent Recognition: Distinguishes between factual queries (listing characters) and analytical requests (auditing pacing or themes).
  • High-Performance Vector Store: Utilizes FAISS (Facebook AI Similarity Search) for sub-second retrieval across large text corpora.
  • 2026-Ready Stack: Built on the latest Gemini 3 Flash Preview architecture for superior zero-shot reasoning.

🛠 Tech Stack

Component Technology
Language Python 3.10+
LLM Google Gemini 3 Flash Preview
Orchestration LangChain (LCEL)
Vector Database FAISS (CPU)
UI Framework Streamlit
PDF Processing PyPDF with Metadata Extraction

📂 Project Structure

├── src/
│   ├── app.py              # Streamlit Frontend & Orchestration
│   ├── preprocess.py       # Document Loading & Metadata Chunking
│   └── generator.py        # AI Persona & RAG Chain Logic
├── .env                    # API Key Storage (Ignored by Git)
├── requirements.txt        # Dependency List
└── README.md               # Project Documentation

⚙️ Installation & Setup

Follow these steps to get the environment running locally on your machine.

1. Clone the Repository

git clone [https://github.com/adexxhh/AI-Literary-Question-Generator.git](https://github.com/adexxhh/AI-Literary-Question-Generator.git)
cd AI-Literary-Question-Generator

2. Setup Virtual Environment

Create the environment

python -m venv venv

Activate on macOS/Linux:

source venv/bin/activate

Activate on Windows:

venv\Scripts\activate

3. Install Dependencies:

pip install -r requirements.txt

4. Configure Environment Variables

Create a file named .env in the root directory and add your Google API Key.

Note: Never commit your .env file to version control.

GOOGLE_API_KEY=your_gemini_api_key_here

5. Run the Application

Start the Streamlit server to view the auditor in your browser.

streamlit run src/app.py

Engineering Highlights (For Evaluators)

This project was designed with a focus on data integrity and retrieval precision. Below are the core technical considerations:

Metadata Persistence: Implemented a custom pre-processing pipeline that attaches page-level metadata to document chunks. This ensures the LLM maintains "spatial awareness" of the manuscript, enabling it to cite specific locations rather than just general context.

Robust Extraction Logic: Designed a fail-safe parser to handle the complex JSON/List response formats of the Gemini 3 Preview API, ensuring 99.9% UI stability even when the model returns structured reasoning data.

Context Optimization: Implemented a top-k retrieval strategy ($k=5$) with 2000-token chunks. This balance maximizes reasoning quality for deep thematic analysis while minimizing token latency for a responsive user experience.