MGC SALES ASSISTANT.
Property questions and structured lead scoring in a focused sales workspace.
The idea
MGC’s real estate sales team needed reliable answers across project documents and a better way to decide which leads to contact first. I brought document assistance and lead prioritization into one lightweight local application.
The challenge
Keep answers grounded in supplied documents and avoid data leakage in lead scoring. The assistant resolves conflicting information using document dates; model training excludes identifiers and information unavailable when a lead first arrives.
The system
The document assistant uses brochures, price lists and booking policies, showing sources alongside each response. It calculates cumulative pricing premiums and avoids unsupported details. Beside it, a logistic regression model scores leads using patterns learned from 9,000 deduplicated historical records.
What I built
Built a responsive, MGC-branded dashboard with Python and FastAPI, Jinja templates, and a simple local setup. The application combines document questions and answers with a structured lead-scoring form, supported by pandas and scikit-learn.
Tools
Python / FastAPI / Jinja / pandas / scikit-learn
The outcome
The lead model achieved an Average Precision of 0.156 against a 0.069 baseline, measuring how effectively it ranks relevant leads. All 19 project tests passed, covering key document answers, scoring and application behavior.
Gallery
