Agrowbot

Lightweight retrieval-augmented QA helper for rice pest control guidance using FAISS and OpenAI chat completions.

Features

• Builds embeddings from local PDF references with all-MiniLM-L6-v2 and stores them in FAISS.
• Serves concise, context-grounded answers via OpenAI chat completions.
• Robust voice interaction loop for interactive questioning.

Project structure

• agri.index, chunks.pkl: FAISS index and serialized text chunks generated by index.py.
• index.py: PDF ingestion, chunking (500-word windows), embedding, and index build script.
• app.py: Runtime retrieval and Q&A loop backed by OpenAI with per-session chat memory.
• requirements.txt: Python dependencies.

Prerequisites

• Python 3.10+ recommended.
• OpenAI API key (set as OPENAI_API_KEY). Optional OPENAI_MODEL overrides the default model.
• Source PDFs placed alongside the scripts (filenames in index.py).

Setup

1. Create and activate a virtual environment.
2. Install deps: pip install -r requirements.txt.
3. Create a .env file in the project root:

   OPENAI_API_KEY=your-key-here

Build the index

Run after placing the PDF files listed in index.py in the project root:

python index.py

Artifacts produced:

• agri.index (FAISS index)
• chunks.pkl (serialized text chunks)

Run the Q&A loop

python app.py

Type a question; enter exit or press Enter on an empty line to quit.

Notes

• Responses strip asterisks and return 2-3 concise bullet phrases when context is available.
• If the question is off-topic, the app replies with: I am sorry, I don't have the information to answer that question.
• Retrieval uses top-3 chunks; adjust k in retrieve() if you need more or fewer.

to activate virtual env .\venv\Scripts\activate