SheetSafe AI - Privacy-First Spreadsheet Analyst

SheetSafe AI is Team Titan's privacy-first spreadsheet analysis MVP, designed for people who need useful answers from CSV files without writing Python code or complicated formulas. Traditional AI spreadsheet tools may depend on executing model-generated code or transmitting entire datasets. SheetSafe AI separates AI planning from local computation to reduce those risks and make each step more understandable. Our application uses a five-page Streamlit interface and a FastAPI control plane running on Vultr Cloud Compute. Users upload a trusted CSV file and receive column profiling, missing-value statistics, summaries, and charts. They can then ask questions in natural language. The backend sends a sanitized question and permitted schema metadata, rather than raw rows or cell values, to Vultr Serverless Inference using the deepseek-v4.1-flash-1017 model. Its structured JSON plan is validated before approved local pandas operations calculate the answer. In a verified browser demonstration, SheetSafe AI identified Student_ID STU_2026244 with the highest age value of 26.00 and displayed the LIVE Vultr inference badge, calculated metric, and summary table. The native Ubuntu 24.04 test suite recorded 1,610 passes and one skip. The MVP includes session controls, CSRF protection, private storage, and session revocation. Arbitrary AI-generated Python or shell code is not executed. Public multi-tenant untrusted uploads and autonomous sandboxed execution remain gated pending additional security acceptance. The current interactive demo is private, accessed through an SSH tunnel rather than a public HTTPS URL.