Mason Lancellotti

Prediction markets have been the subject of my fascination for multiple years at this point, since they include such a wide, near-ridiculous array of subjects and events that you can “invest” in. It seemed logical to me that since these markets are still new (compared to other, more established financial markets) and attract an abundance of uninformed retail traders, their potential to contain inefficiencies would be high.

Crypto Contract Backtesting System

This project involved short-term crypto contracts on Kalshi, featuring leakage-safe data handling, purged cross-validation, probability calibration, fee-aware backtesting, and multiple checks against a sealed holdout. 38 strategies, all generated from a data mining-based discovery engine, can be tested, which I did across millions of real trades and just shy of 700 iterations. Small edges were found here and there, but most were unreliable or consumed by fees. In the end, I discovered that the market is indeed efficient. I would like to use this program on other exchanges in the future.

Arbitrage Scanner

This idea is what originally made me interested in prediction markets; since there are so many venues, there must be price differences between them for the same contracts, even if only for a few seconds. In this project, I combined graph-based relationship detection and caching, fee-specific edge math, and a layered verification stack with LLMs classifying how two markets or contracts relate, with hard-coded validators preventing them from being overwhelmed or wasting tokens. On a live run, the program passed over 150,000 contracts and 300,000 quotes, identifying 800 pairs and rejecting all of them due to rule mismatches or incomplete outcome sets. In the future, I would like to expand the number of venues this program supports, as well as improve its process and cost so that it can run continuously and more efficiently/accurately identify and potentially autonomously profit off of arbitrage opportunities.

Trading Terminal

A Bloomberg-style four-panel desktop app wired to live public data from both Kalshi and Polymarket. The interface includes function keys, full keyboard navigation, most-active and biggest-mover screens, news, charts, order book depth, live latency and connection health indicators, a scrolling ticker tape, world clocks, a watchlist, and a local paper-trading environment that fills simulated orders against the real book. One challenge I was proud of tackling in this project is how Kalshi and Polymarket return data in completely different ways; on the backend, all data is converted into one common format, so contracts from both exchanges can sit side by side in the same list and be compared directly.

Weather Contract Pricing Model

This project involved daily high-temperature contracts on Kalshi and Polymarket. It reads contract fine print to determine the exact weather station or sensor a market will settle on and then calculates a machine learning-based fair-value price based on that station's NWS data, building a baseline from recent daily highs, blending in the current forecast, and using historical forecast error to set the spread of the distribution. Because many of these contracts settle on a range rather than a single temperature, it integrates that distribution across each bucket to yield a probability per outcome, which it then conveniently compares against the order book (with prices spanning from zero to one and summing to one, exactly like a probability) to determine edge after fees. In the future, I would like to improve/expand this program’s pricing methodology, as well as create a more concrete, automated trading mechanism.

A custom application built for GROWTH Homes that includes the end-to-end path for turning raw property footage into scheduled social media content. It was made with real estate development in mind but could also be generalized for use in any subset of the real estate industry. The app is highly AI driven, with a multitude of APIs making it a one-stop shop for strategy, video storage, content production, post scheduling, and analytics. It also features company- and competitor-based web and social scraping, a notebook that natively integrates into video creation workflows and strategy development, an AI “consultant” trained on 40+ pages of highly specific industry context and reference files, cloud-based media storage, and functional team-based features. Across a pilot batch of seventy-six publication-ready videos it saved an estimated sixty-one hours of manual work. There is no public repository for this project, as it belongs to my employer. I plan to continue working on this project seriously, making improvements and eventually adding new features.

A local research system that runs continuously and produces scheduled market briefs as well as ad-hoc investment briefs. It ingests equities, crypto, news, SEC filings, macroeconomic data, and prediction markets, computing a composite risk regime across them, and running an agent pipeline, including data volume filtration, analysts, adversarial review, a dedicated material composition stage, and a fact-checker. This process yields cited morning digests, strategy notes, and hedge ideas. My main focus was the review stage, which checks the brief’s claims against sources and challenges weak assumptions.

Evaluates listings on secondhand markets against recent comparable sales and handles the full transaction workflow if a listing is identified as likely to be profitable. From discovery and valuation through communicating with sellers, purchasing, listing, repricing, and shipping, every action is recorded, and the level of automation is highly customizable by user. The program is also self-training, following the listings it declined through to their eventual sale, scoring valuation error and confidence to tune the opportunity identification system over time.

A portfolio website built with HTML, CSS, and JavaScript to give a fuller picture of my background than a resume alone. It brings together work experience, coursework, project writeups, and market articles, with layouts for desktop, tablet, and phone screens. The Skills Index opens relevant examples in a preview so readers can explore the work without losing their place.