Show HN: Web client analyzing prediction market outcomes
o-u.aiHi HN,
I made a web client that analyzes prediction markets. Please send your critical feedback to a struggling solo dev.
Functionality:
1. 2hr cron query to Kalshi/Polymarket API for current market pricing and contract data.
2. Query relevant data APIs - 12 right now: NWS, congress.gov, Odds API...
3. Serper/Tavily search on each contract. These APIs pull relevant search results, summarize, and send it back to you in JSON.
4. Structured call to Gemini/Sonnet for each contract or event series containing multiple contracts. There are around 9k active analyses, but each cron only runs price moves or un-assessed contracts. Lots of backstops to prevent analysis runs getting stuck.
5. Store in Postgres database.
6. If a contract/event isn't new and doesn't have a price move, batch submission to Gemini on a 24hr cron to prevent staleness.
7. Bank account balance goes down.
Outlook:
- Good success with lower volume, lower scrutiny contracts.
- Still testing for accuracy but it's going well. Can't find an edge on major sport events because they are so heavily traded by professionals.
- Predictably can't get an edge on high volume contracts where Susquehanna is running a full quant stack.
- Can't compete with retail quant bots on timing due to refresh restrictions. Could be useful for market making. What have you tried doing with it? If you loaded the analysis pipeline on a bot and stuck to certain contracts it might be half decent at market making low volume contracts. The quant specialists reportedly ignore them. I have been working on an idea to buy a large basket of contracts since the advantage of having quick access to all the analyses is breadth, and if your model beats the market consistently you can participate low risk/low impact.