It is a dashboard for providing deep statistical analysis for financial records, paired with agentic tool calling via Chat interface, with the ability to export analysis artifacts in multiple format
Inspiration Construction contractors lose millions in margin erosion because financial data is buried across spreadsheets, billing apps, and field notes. We wanted to give project managers an AI copilot that instantly surfaces where money is being lost and why.
What it does HVAC Deep Data is an AI-powered construction portfolio analytics platform. It lets users chat in natural language to analyze profit margins, labor productivity, billing status, RFIs, and change orders across 5 HVAC projects — backed by real construction data (16K+ labor records, 300+ RFIs, 60+ change orders).
How we built it Next.js + React frontend, SQLite database seeded from CSV construction datasets, Anthropic Claude via the Vercel AI SDK for the chat interface, and 22 specialized tool calls that let the AI query financial, labor, risk, and portfolio data in real time. We also built an MCP server for Claude Desktop integration.
Challenges we ran into Deploying SQLite on Vercel's read-only filesystem required copying the DB to tmp at runtime. Aggregating burdened labor costs with overtime premiums across 16K records needed careful SQL to stay performant. Designing tool calls that give the AI enough context without overwhelming the context window was a constant balancing act.
Accomplishments that we're proud of 22 AI tool calls that cover the full spectrum of construction financial analysis — margin erosion, labor productivity, RFI risk scoring, cross-project pattern detection, and field note search. The AI can pinpoint exactly which SOV line is bleeding money and back it up with data.
What we learned How to design effective AI tool schemas for domain-specific data, the nuances of construction billing (retention, SOV, AIA pay applications), and that giving an LLM structured access to a relational database is far more powerful than RAG over documents.
What's next for HVAC Deep Data Real-time integrations with Procore/Sage, predictive margin forecasting using historical trends, automated early-warning alerts for at-risk SOV lines, and expanding beyond HVAC to general commercial construction.
Log in or sign up for Devpost to join the conversation.