SentryQuery-AI

A minimal Retrieval-Augmented Generation (RAG) prototype for asking questions over a folder of PDFs. It indexes documents into Pinecone and uses LangChain (LCEL) + OpenAI to generate answers grounded in retrieved context.

Features

• Loads PDFs from ./docs/
• Chunks text for retrieval
• Creates embeddings with OpenAI
• Stores/retrieves vectors with Pinecone
• Builds a context-grounded prompt and answers with GPT via LangChain

Architecture (Workflow)

Ingest (Indexing)

1. Load environment variables from .env
2. Load PDFs from ./docs
3. Split text into overlapping chunks
4. Embed chunks
5. Upsert embeddings to Pinecone index (sentry-index)

Query (Retrieval + Generation)

1. Take a user question
2. Retrieve top-matching chunks from Pinecone
3. Insert chunks + question into a prompt template
4. Generate an answer with the chat model
5. Print the result

Requirements

• Python 3.11+
• An OpenAI API key
• A Pinecone API key (and an existing Pinecone project)

Setup

python3 -m venv venv
source venv/bin/activate

pip install -U langchain langchain-community langchain-openai langchain-pinecone pypdf pinecone-client python-dotenv

Configuration

In the project root (next to sentry_query.py), create a file named .env. It is gitignored; do not commit it.

Add exactly these two variables, each on its own line, with your real keys after the =:

OPENAI_API_KEY=your_openai_key_here
PINECONE_API_KEY=your_pinecone_key_here

load_dotenv() in sentry_query.py loads this file. LangChain’s OpenAI classes read OPENAI_API_KEY. Pinecone reads PINECONE_API_KEY.

Add Documents

Place one or more PDFs into:

• ./docs/

Run

./venv/bin/python sentry_query.py

Example Queries

Edit the query = "..." line in sentry_query.py and try:

• "Summarize the main purpose of these documents."
• "List key requirements or controls and group them by category."
• "Extract an implementation checklist."
• "What access control or PII-related guidance is described?"

Notes / Limitations

• This script currently re-embeds and re-upserts on every run (good for a demo, inefficient for production).
• Data is sent to cloud services (OpenAI + Pinecone). Avoid indexing confidential data unless you have permission and governance controls.

License

MIT (see LICENSE).