RAG'ed Mankuthimmana Kagga GPT 📜

Your Personal Philosopher:This project is a Retrieval-Augmented Generation (RAG) system that uses a Large Language Model (LLM) to provide philosophical insights from Mankuthimmana Kagga. It leverages a curated knowledge base to ground the LLM's responses, making them contextually accurate and relevant.

---

Tech Stack for my experiments with LLM augmentation using RAG, vector embeddings, and vector DB store involves:

• Agent Framework: LangChain for orchestrating the RAG workflow.
• Vector Embeddings: A Google embedding model to convert text into vector representations.
• Vector Database: Pinecone for high-performance storage and retrieval of embeddings. Dense with 768 dimensions insights.
• Generative Model: Gemini on LangChain for generating coherent responses.
• Language: TypeScript for a robust and type-safe application.

---

How It Works:The system follows a standard RAG pipeline:

• Unstructured data source from pdf, docs, etc. is loaded, chunked, and embedded using Langchain tools and Google embedding model.
• These embeddings are stored in a Pinecone vector database and indexed.
• When a user query is received, the system retrieves the most relevant text chunks from the database.
• The retrieved context is then provided to the Gemini LLM, which uses this information to generate a grounded and insightful response, ergo ur Philosopher.

---

Getting Started Prerequisites:

• Get your Pinecone API Key and Index.

• Google API key.

• Installation:

git clone <repository-url>

npm install -g bun && bun install

cd src

• Configuration: Create a .env file inside src folder and add your keys, example below:

PINECONE_API_KEY=<YOUR_PINECONE_API_KEY>
GOOGLE_API_KEY=<YOUR_GOOGLE_API_KEY>
PINECONE_INDEX = <YOUR_PINECONE_INDEX>
DO_EMBED_COSTLY_OP = <ENABLED | DISABLED>

Note!: You should be in src folder by now, to run these cmds:

Note!: Since, ingestion is a costly ops! feature flag has been provided below to avoid any accidental runs!

• To populate the vector database with embeddings (manual upload):

in your .env file set below feature flag to enable the ingestion:

DO_EMBED_COSTLY_OP: ENABLED

then:

bun run ingest.ts

Note!: You may want to turn back the above feature flag to DISABLED now!

• Finally To run chat:

bun run philosopherChat.ts 

• Once you are done with your query type /bye to exit app.

---

Acknowledgments: Although this project is for my understanding of RAG-based augmentation, it's also a tribute to D. V. Gundappa, whose timeless wisdom serves as a cornerstone for many. His work continues to inspire generations.

Humbled AI:

Trying to bring code and art-storytelling together.