Retrieval-Augmented Generation (RAG) Portfolio

Project 1: Enterprise-Grade RAG Chatbot Analysis

This architecture was reverse-engineered from a live e-commerce application (Jason's Cookie Company). It demonstrates a robust, scalable, and secure **Serverless RAG** pattern utilizing a dedicated vector database.

1. Frontend

User submits question via HTML form.

↓→

2. API Gateway

https://49dyop31n3.../ask

Entry point, routes request securely.

↓→

3. AWS Lambda (Orchestrator)

Code executes RAG logic, contains **secure API Key** to Pinecone.

⬇ Query / ⬆ Context↔ Retrieve Context

4. Vector Database

**Pinecone Index: `chatbot`**

Host: `chatbot-6dtmms8...pinecone.io`

Model: `text-embedding-3-small`

⬇ Context / ⬆ Answer↔ Send to LLM

5. Large Language Model (LLM)

Generates final answer based on retrieved context.

Key Technical Takeaways:

**Serverless Architecture:** Utilizes **AWS API Gateway** and **AWS Lambda** for cost-efficient, auto-scaling backend.
**Dedicated Vector Store:** Confirmed use of **Pinecone** for high-performance semantic search, demonstrating an understanding of specialized ML infrastructure.
**Security:** Critically, the **Pinecone API Key** is hidden in the Lambda Environment Variables, ensuring data security—a crucial enterprise requirement.

Project 2: Basic Gemini RAG Chatbot (Your Implementation)

This simple RAG implementation uses a lightweight approach for a personal knowledge base, demonstrating the fundamental RAG concept without the complexity of a full vector store.

1. Frontend

User inputs query.

↓→

2. Gemini API / Tool

The LLM is provided with the custom knowledge file as a **Tool** or **Context**. Gemini searches the file directly.

⬇ Read File↔ Access Knowledge

3. Knowledge Base

aboutme.txt

Simple unstructured text file (e.g., in a local directory or S3 bucket).

Key Learning Points:

**Core RAG Principle:** Successfully implemented **Retrieval** (scanning the text file) and **Generation** (using Gemini) to ground the response.
**Cost Efficiency:** Demonstrates the ability to build a functional, low-cost RAG system for a defined, small knowledge scope.
**Conceptual Understanding:** A clear understanding of how to provide **external context** to an LLM to prevent hallucinations and provide accurate, specific answers.