Project Portfolio

Creating the Agent

This was a fun two-day agent project to build, inspired by something I wish I had had when I ran my digital agency. It’s so easy to forget about clients when they don’t e-mail very often, right? 

This multi-function CRM built on Fabric, Python and Athropic AI aims to fix that.

First, it will scan your inbox data for the latest updates every morning about your clients and their campaigns. For those clients you haven’t heard from in 20-30-60 days, an automatic email will be generated and sent to you to remind you to reach out:

But before you send that e-mail or jump on a call, be sure to brush up on anything that you’ve talked about in communications via e-mail in the E-mail Assistant (in-browser) which scans every e-mail from them for information that you probably don’t remember – and reports it back to you based on your chat questions:

Finally, to make sure you know everything about their website’s latest performance, jump into a Fabric notebook for the most important website analytics’ updates, along with custom AI recommendations, powered by Claude.

Et voila! Your business is generating more leads – and taking better care of your clients – overnight.

Project Specs

Data Engineering Layer

Tools: Zoho Mail + Google Analytics + Microsoft Fabric

1. Raw Email data mining: I began by using Microsoft Fabric and Zoho API to pull raw email data into the database. Then, I transformed the data with Python into clean, readable data.
2. Google Analytics data mining: Next, I pulled raw data from Google Analytics into Fabric using DataFlow Gen2.
3. Data Harmony: Finally, I created a join table to match the clients and their emails to their respective GA4 data tables via common domains.
4. Daily Refresh: Each layer is combined in the Fabric data pipeline to update every morning before I turn on my computer, in a sequence that makes sense for data integrity.

Front-End Layer: Custom CRM  + AI magic!

Tools: Anthropic AI (Claude) + Python + Streamlit + E-mail + Fabric Notebooks

1. Automated Inactivity Alerts — runs every morning at 6am, checks both directions of email contact across all clients, cross-references their web traffic, and emails your team a formatted report of anyone who’s gone quiet.
2. Email Chatbot — a local Streamlit app that lets anyone on your team ask plain English questions about your client email history and get intelligent answers backed by semantic search.
3. Website Insights — a Fabric notebook that pulls real GA4 data per client, calculates traffic trends and drops, visualizes their session history, and generates specific AI-powered improvement suggestions.