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.