Projects
Building tools that solve real problems
Real Estate Opportunity Engine
Mar 2026
A full stack platform that scrapes Florida MLS listings, trains an ML model to predict new build values, and scores every property by how profitable a knock-down rebuild would be.
Key Features
- ▪XGBoost model trained on MLS sales data with a 0.5-mile hyper-local price feature
- ▪Supports four ML algorithms (XGBoost, Random Forest, Ridge, LightGBM) with live R² comparison
- ▪React-Leaflet map with color-coded opportunity scores and comparable sales within 0.5 miles
- ▪Live operations console for kicking off scrape and training jobs from the browser
- ▪Full Docker Compose stack with Postgres named volume and Caddy HTTPS
Combines a Python data engine (HomeHarvest scraper, XGBoost/Random Forest/Ridge/LightGBM), a Node.js API, and a React dashboard to identify buy-demolish-rebuild opportunities across Tampa, Orlando, Winter Garden, and Winter Park. Computes a hyper-local price baseline using a 0.5-mile Haversine radius around each property. Opportunity score = predicted rebuild value minus acquisition cost minus construction cost. Runs entirely in Docker Compose with Postgres persistence and a Caddy reverse proxy.
Triple-Reasoning Chat Analyzer
Feb 2026
Taking an AI Ethics class got me thinking about how AI actually reasons. A chat app that responds to every question three times, deductively, inductively, and abductively, to make the difference visible.
Key Features
- ▪Three parallel LLM calls per message with distinct reasoning prompts
- ▪Deductive, inductive, and abductive response structures enforced via prompt engineering
- ▪Logic analysis mode: symbolic form, premises, conclusion, fallacy detection
- ▪Response structure guidelines so users can evaluate model compliance
- ▪Dockerized with nginx reverse proxy and HTTPS via Let's Encrypt
Built to explore ideas from IDH3600 (Professor Prevaux) around AI reasoning and critical thinking. Sends each user message to an LLM three times in parallel, each with a prompt that forces a distinct reasoning structure. Includes a logic analysis feature that extracts symbolic form, argument structure, and fallacy detection. Built with React, Node.js, Docker, and Llama 3.3 70B via OpenRouter. Deployed at aiethics.carlostrevisan.xyz.
AI Email Summarizer
Jan 2026
A containerized email summarization system that processes overnight emails using AI, providing concise daily digests with smart prioritization. Built with Node.js, Docker, and Llama 3.3 70B via OpenRouter.
Key Features
- ▪Automated IMAP-based email fetching and parsing
- ▪AI-powered summarization with conservative urgency detection
- ▪Smart email clustering to reduce redundancy
- ▪HTML digest generation with web archive
- ▪Containerized deployment with scheduled execution
Automated email intelligence platform that connects to Gmail via IMAP, processes unread messages through an LLM, and generates prioritized HTML summaries. Features smart email clustering, conservative urgency detection, and a web archive for historical summaries. Runs as a scheduled Docker container with customizable digest delivery.
