Cost Variance / Profitability Analysis Tool

Construction firms struggle with post-project analysis, spending hours manually comparing estimates to actual costs in Excel spreadsheets. The process is time-consuming, error-prone, and lacks intelligent insights. Project managers review variance reports but miss patterns across projects, while executives lack narrative explanations for budget overruns. Historical project data sits unused, preventing learning from past mistakes. The result: reactive analysis, missed optimization opportunities, inconsistent reporting, and limited strategic insights for future project planning.

I designed an AI-powered cost variance analysis system that transforms raw Excel data into intelligent, narrative-driven insights using Google Cloud's Vertex AI Gemini 1.5 Pro. The system automatically parses estimate and actual cost files, calculates comprehensive variance analysis, and generates executive-level explanations with actionable recommendations. Built-in memory storage enables pattern detection across projects, while a beautiful web interface makes the tool accessible to non-technical users.

• System Architecture — Designed modular pipeline around Google Cloud native services.
• AI Integration — Implemented Gemini 1.5 Pro for financial analysis with structured prompt engineering
• Data Processing — Built flexible Excel parser with auto-column detection and variance calculation
• Memory System — Created Firestore-based project memory with embeddings for pattern detection
• Web Interface — Developed Flask-based web UI with Tailwind CSS for non-technical users
• Deployment — Configured Cloud Run deployment with Docker containerization

This demonstrates translating complex financial analysis into an intelligent, scalable system that balances automation with explainability.

Excel Format Variability: Different firms use varying column names and structures, requiring flexible parsing logic with auto-detection capabilities
AI Prompt Engineering: Financial analysis requires nuanced prompts to generate executive-level insights rather than generic summaries
Memory Architecture: Firestore lacks native vector search, requiring custom embedding storage for pattern detection
Cost Optimization: Vertex AI pricing requires careful token management and fallback strategies for cost-conscious deployment
User Experience: Non-technical users need intuitive interfaces that hide complexity while maintaining analytical depth