Portfolio

Building a Deterministic Lead Qualification API for Salesforce Lead qualification sounds simple: receive a business name and address, look up the company, and create a Lead in Salesforce. The difficult part starts when the same restaurant arrives from two sources, the address is written differently, a search returns several similar businesses, or two requests reach the API at the same time. A system that guesses incorrectly can enrich the wrong company. A system that checks for duplicates too late can create two Salesforce records. A system without an audit trail cannot explain why a lead was accepted or rejected. ...

Testing AI Agents with Evals AI agents are probabilistic. The wording can change between runs, the agent may take different paths through its tools, and a slightly different response doesn’t necessarily mean the agent made a wrong decision. This mmakes it challenging for traditional software testing to produce reliable and actionable results and findings. Using an LLM to judge every test isn’t the answer either. It adds latency, cost, and another source of variability to the test suite. ...

Finding Lead and Account Duplicates with Salesforce Agent Script Duplicate data is one of those CRM problems that is frustrating and can slow down the sales team. For this project, I built a Salesforce Agent Script agent that helps answer a very practical question: How confident are we that this Lead represents this Account? Instead of letting the agent guess, I connected it to deterministic Apex actions. The agent can have the conversation, but the scoring logic comes from Salesforce data and clear rules. ...

Enriching Salesforce Leads with Place IDs I built a Python-driven automation that enriches Salesforce Leads using Google Place IDs as a stable identifier for businesses on Google Maps. The goal was to move beyond manual lead research. If a Lead already has a Place ID, the script can use that identifier to fetch structured business details and sync the useful fields back into Salesforce. The Three Ingredients 1. Salesforce Lead Object Configuration To prepare Salesforce for enrichment, I added custom fields to the Lead object. This included a Place ID field to serve as the lookup key, along with fields for Google rating, review count, website, and other useful business details. ...

Searching for Leads with Salesforce Agentforce Building on my lead enrichment automation, I wanted to create a more intuitive way for sellers to interact with their data. Instead of scrolling through list views, I built a custom Agentforce Agent that allows sellers to find high-value prospects simply by asking. The Three Ingredients 1. Custom Fields & Logic To make the search meaningful, I added industry-specific fields like Estimated Monthly Orders, Average Order Value, Cuisine, and delivery availability. I then built a Lead Score Flow that calculates a priority score using this formula: ...

Basic Lead Enrichment in Salesforce with Agentforce To get more hands-on experience with Salesforce automation, I used a Developer Edition org to build a basic AI-assisted lead enrichment flow. When a new Lead is created with limited company information, the automation uses Agentforce to look for missing contact details and then updates the Lead record. The goal was simple: reduce manual research for fields like phone number, email, and website, while still keeping the setup easy to inspect and improve. ...

How Meal Recommendations Work Imagine you’re running a food marketplace with thousands of users and hundreds of meal options. How do you figure out what someone might want to eat? Two people who look identical on paper might have completely different taste in food. One person loves spicy Thai dishes, the other can’t stand them. So if demographics aren’t the answer, what is? Find People Like You Collaborative filtering works on one simple idea: ...

AI-Enhanced Monitoring with Grafana and CrewAI Monitoring systems have come a long way in helping teams keep an eye on infrastructure. Tools like Prometheus, Grafana, and Alertmanager are great at collecting metrics, showing dashboards, and sending notifications when something crosses a threshold. But alerts often stop at something is wrong. They do not always explain why it happened, what changed recently, or what the next step should be. To explore that gap, I built a small proof of concept that connects Prometheus, Alertmanager, Grafana, and CrewAI. The idea was not to replace the monitoring stack, but to add an AI layer that can interpret alerts, add context, and produce a more useful incident summary. ...

AI-Powered Freelancer Payout Crew This project is a multi-agent payout workflow built with CrewAI. It simulates the lifecycle of freelancer payments, from validating payout data to checking compliance rules, selecting a payment route, preparing the payment instruction, and creating an audit trail. The goal was to explore how agent-based workflows can make a finance process more structured and easier to review. It is a prototype, not a production payment system. The Business Challenge Managing payments for a global freelance workforce is a complex, manual, and time-consuming operation. Finance teams are burdened with: ...

Resume Checker - Chrome Extension In today’s job market, it is difficult to know whether your resume really matches a specific job description. I built a Chrome extension to make that comparison faster and more practical. The tool gives job seekers immediate feedback by comparing their resume against a job description, highlighting matching keywords, missing keywords, and concrete improvement areas. What I Built The Resume Checker lets users analyze a job description directly from the browser. They paste their resume text, run the check, and get feedback without leaving the page they are viewing. ...