Fitness RAG Bot with Amazon Bedrock AgentCore

A Retrieval-Augmented Generation (RAG) chatbot deployed on Amazon Bedrock AgentCore that answers fitness-related questions using a knowledge base and remembers conversation history across sessions.

The bot: - Retrieves fitness knowledge using FAISS vector search - Uses Groq LLM (Llama‑3.3‑70B) for responses - Stores conversation history using AgentCore Memory - Runs inside Amazon Bedrock AgentCore runtime

---

Architecture

User Query ↓ AgentCore Runtime ↓ Memory Retrieval (AgentCoreMemoryStore) ↓ RAG Retrieval (FAISS Vector Store) ↓ LLM Generation (Groq / Llama‑3.3‑70B) ↓ Store Conversation in AgentCore Memory

---

Features

• Retrieval Augmented Generation (RAG)
• Persistent memory across sessions
• Amazon Bedrock AgentCore deployment
• Groq LLM integration
• FAISS vector database
• LangChain based tools

---

Project Structure

.
├── main.py
├── rag_tool.py
├── vector_store.py
├── documents/
├── pyproject.toml
└── README.md

---

Prerequisites

You need:

• AWS Account
• AWS CLI configured
• Python 3.12+
• AgentCore CLI installed
• Docker installed
• Groq API Key

---

Environment Setup

Create environment variables:

Linux / Mac

export GROQ_API_KEY=your_key

Windows

set GROQ_API_KEY=your_key

---

Install Dependencies

pip install -r requirements.txt

or if using pyproject.toml

pip install .

---

Build Vector Store

Run the script that loads documents and builds FAISS index.

Example:

python vector_store.py

This creates a FAISS index used for retrieval.

---

Run Locally

python main.py

Or using uvicorn if FastAPI is used:

uvicorn main:app --reload

---

Deploy to Amazon Bedrock AgentCore

1. Login to AWS

aws configure

Set:

• Access Key
• Secret Key
• Region

---

2. Initialize AgentCore

Inside the project directory:

agentcore init

This generates the agentcore.yaml configuration file.

---

3. Configure Agent

Edit the configuration file:

entrypoint: main:app
language: python
deployment_type: container
platform: linux/arm64

Add your IAM execution role.

---

4. Deploy Agent

agentcore deploy

This step will:

• Build a container
• Push image to Amazon ECR
• Create AgentCore runtime
• Deploy the agent

---

5. Set Environment Variables

Example:

agentcore deploy --env GROQ_API_KEY=your_key

---

Invoke the Agent

Run:

agentcore invoke "{'input':'How much water should I drink daily?'}"

Example response:

{
 "answer": "The general recommendation is to drink 2–3 liters of water daily."
}

---

Conversation Memory

The bot stores messages using:

AgentCoreMemoryStore

Each session is stored using:

(actor_id, thread_id)

This allows the bot to remember previous questions within the same session.

Example:

User: How much water should I drink daily?
Assistant: 2‑3 liters

User: What did I ask earlier?
Assistant: You asked about daily water intake.

---

Useful Commands

Deploy

agentcore deploy

Invoke

agentcore invoke "{'input':'your question'}"

View logs

aws logs tail /aws/bedrock-agentcore/runtimes/<runtime-name> --follow

---

Tech Stack

• Amazon Bedrock AgentCore
• LangChain
• LangGraph Memory
• FAISS
• Groq LLM
• Python

---

Future Improvements

• Multi‑agent architecture
• User preference memory
• Streaming responses
• API gateway integration
• Web UI

---

Author

Aditya Naranje

AI / ML Engineer | Generative AI | LLM Applications