Our biggest inspiration is the rise of predictive markets. Predictive markets allow everyday people to get a sense of investing in an easy accessible way. We wanted to expand that accessibility to another big aspect of finance: hedge investments. Traditionally, hedge investing requires complicated financial instruments: derivatives, shorts, and non-correlated assets to offset potential losses. This is not something the everyday person can easily master and apply to protect their personal portfolio. So we built Offset to simplify that process for everyday investors, so they can also use hedging to control their risks and losses.
Offset is an agentic hedging bot that connects to your brokerage portfolio via Plaid, identifies the specific risks most likely to hurt a position you are worried about, and automatically finds Kalshi prediction markets that pay out if those risks materialize. Instead of selling your shares and losing your upside, you place a small targeted trade for the exact event you are scared of. Offset handles the research and filtering of hundreds of prediction markets, presenting you with the best markets to be used to hedge. All the user has to do is connect their portfolio and tell the chat interface what investment they want to hedge.
The core of Offset is a multi-stage AI pipeline that takes a stock ticker and produces a ranked list of relevant Kalshi prediction markets. The pipeline first decomposes the stock into its key risk factors, generates targeted search queries for each risk, retrieves candidate markets from the Kalshi API, scores them for hedge relevance, and finally generates a plain-English hedge thesis for each recommendation. To wrap this algorithm, we built a conversational LLM-powered chat interface which performs tool calling, so users can buy, sell, and research hedges entirely through chat messages. Portfolio data is pulled from Plaid's investments API, all state is persisted in Convex, and the frontend is built with Next.js, Tailwind, and Lovable. The chat interface uses OpenAI API and the hedging bot uses OpenAI API with web search and Kalshi API.
The hardest problem we faced was identifying hedge relationships that are not immediately obvious. A direct hedge is easy - if you own an oil company, look for oil price markets. But there are many viable hedges that are indirect. An investment in gold, for example, can be effectively hedged with a prediction market about conflict in a major gold-producing nation, because that event would move gold prices meaningfully. Instructing the AI pipeline to discover and reason about these non-obvious causal chains and then rank them by the strength of their relationship to the underlying investment required significant iteration on our prompting approach and pipeline structure. In total, a single call for hedges will require several iterations of LLM inferences + web searches that are mutated and used as input for more inferences, in addition to Kalshi API fetch calls.
We are proud of the multi-step hedge discovery pipeline, which surfaces direct and indirect connections between investments and prediction markets that an ordinary investor would be unlikely to find on their own. The chat-first interface ended up being cleaner than we expected - having every action flow through a single conversational input + tool call rather than scattered UI controls that would have made the product feel approachable and scary. The payment system we set up is another major part of our project which facilitates payment on the user’s behalf using Blockchain technology ensuring a secure payment system. We are also proud of how well the team came together to devise a project plan/design documents and divided the work across major parts of the overall project.
We realized the untapped potential to use prediction markets as a hedging instrument for everyday investments. From the limited amount of time we had to build this project, we were able to devise an algorithm that took security as input and discovered several strong and sensible hedging relationships against various Kalshi markets. We also learned a lot about the practical limits of LLMs in financial reasoning. They are excellent at decomposing risk qualitatively but need to be carefully structured to avoid producing confident-sounding but unusable data.
The next thing on our roadmap is making Offset fully autonomous. Right now, the pipeline identifies strong hedge candidates and presents them to the user, who then decides whether to buy or sell. We feel that is still too much work on the user's end. The vision is for Offset to asynchronously monitor every open position in the background, tracking how the underlying market odds are shifting in real time and autonomously deciding when to buy, when to hold, and when to sell in order to optimize hedges against your complete portfolio. You connect your portfolio, tell Offset what you are worried about, and Offset takes it from there, no extra worrying required.