E-Commerce Personal Shopper

Given an e-commerce shopping related query, the E-Commerce Personal Shopper assistant finds products, compares prices, and analyzes reviews using real-time scraped data. And then provides personalized suggestion.

For example:

If the user asks, "I want an ipad under ₹65,000 with a great camera and good battery. I prefer Samsung."

Then,
Remote_Agent_1 [Product Recommender Agent]: Recommends products for the user query

Remote_Agent_2 [Price Tracker Agent]: Scrapes Amazon website [MCP tool] and fetches the purchase links and prices for the products.

Remote_Agent_3 [Review Analysis Agent]: Uses google_search and analyzes the reviews for the products suggested by the 'Product Recommender Agent'.

E-Commerce Personal Shopper: Finally, provides personalized suggestion based on the remote_agents analysis.

The E-Commerce Personal Shopper acts as a root agent [GoogleADK]. It has three remote agents.

1. Product Recommender Agent [GoogleADK Agent] -> For the user query based on e-commerce, suggests products to purchase. It uses google_search.
2. Price Tracker Agent [CrewAI] -> Uses MCP tool and fetches live prices for products suggested by the Product Recommender Agent. It scrapes the Amazon website.
3. Review Analysis Agent [Google ADK] -> Analyzes reviews for the products suggested by the Product Recommender Agent. It uses google_search

The Personal_Shopper (the root agent) uses SequentialAgent workflow. It first runs the Product Recommender Agent. Then using ParallelAgent workflow, Price Tracker Agent and Review Analysis Agent gets executed in parallel. It takes the input of Product Recommender Agent. Finally E Commerce Personal Shopper provides personalized suggestion for the purchase of the product.

Agents Involved

Agents	Framework	Status
Product Recommender Agent	Crewai	✅ Completed
Price Tracker Agent	Google ADK	✅ Completed
Review Analysis Agent	Google ADK	✅ Completed
E Commerce Personal Shopper (Root Agent)	Google ADK	✅ Completed

What Makes This Project Unique?

1. Connected the root agent with multiple remote agents using RemoteA2aAgent().
2. Implemented a workflow using both SequentialAgent and ParallelAgent architectures.
3. Demonstrated how to use Gemini models.
4. Demonstrated how to use Ollama models.
5. Demonstrated how to use MCP tools.
6. Showcased the use of defaults tools (google_search).
7. Integrated different frameworks, including Crewai and Google ADK.
8. Demonstrated how to connect the root_agent with the remote_agents for collaborative task execution.

Setup

conda create -n capstone_demo python=3.12 -y
conda activate capstone_demo

git clone git@github.com:Jereshea/AgenticAI_CapstoneProject.git
cd AgenticAI_CapstoneProject/requirements/
pip install -r requirements.txt
pip install litellm==1.77.4
curl -fsSL https://ollama.com/install.sh | OLLAMA_VERSION=0.12.1 sh

Models and Deployments

In this example, we use the following models:

1. gemini-2.0-flash (Requires GOOGLE_API_KEY)
2. a custom Ollama model (llama3.1:8b -> llama3.1-cpu-custom)

Gemini Model:

Generate an API key

export GOOGLE_API_KEY= {Your_API_Key}

Ollama model:

CPU:
This demo was originally tested on a CPU-enabled system. If you want to run it on CPU, use the steps below. In many cases, Ollama models actually perform better on a single-socket CPU setup, so CPU execution can be more efficient depending on your hardware.

ollama pull llama3.1:8b
ollama show --modelfile llama3.1:8b > Modelfile_cpu_llama3_1

Add the line 'num_gpu=0' and 'PARAMETER num_thread 128 ' in the file 'Modelfile_cpu_llama3_1'

PARAMETER stop <|start_header_id|>
PARAMETER stop <|end_header_id|>
PARAMETER stop <|eot_id|>
PARAMETER num_gpu 0
PARAMETER num_thread 128

Now create a custom model

sudo ollama create llama3.1-cpu-custom -f Modelfile_cpu_llama3_1

GPU [Optional]:

ollama pull llama3.1:8b  
ollama show --modelfile llama3.1:8b > Modelfile_cpu_llama3_1
sudo ollama create llama3.1-cpu-custom -f Modelfile_cpu_llama3_1

Note: Ollama performs better in single sockect, hence added the line to control threads in the modelfile.

To Run

Terminal 1

conda activate capstone_demo

cd AgenticAI_CapstoneProject/remote_agents/price_tracker

python __main__.py

Terminal 2

conda activate capstone_demo

cd AgenticAI_CapstoneProject/remote_agents/product_recommender

export GOOGLE_API_KEY= {Your_API_Key}

python __main__.py

Terminal 3

conda activate capstone_demo

cd AgenticAI_CapstoneProject/remote_agents/review_analysis

export GOOGLE_API_KEY= {Your_API_Key}

python __main__.py

Terminal 4

conda activate capstone_demo

cd AgenticAI_CapstoneProject

python gradio_script.py