Quant / Prediction Markets
Market Signal Engines
Sentiment pipelines, arbitrage scanners, and prediction-market tooling for crypto and equities.
The problem
Discretionary trading decisions are hard to audit. I wanted the signal generation written down as code — reproducible, testable, and comparable across venues.
How it's built
- Ingestion of price, social, and news data into a Postgres store
- Sentiment scoring pipelines in Python with per-source weighting
- Arbitrage scanner comparing funding and spreads across venues
- Prediction-market probability feeds used as an independent signal against price
- PineScript indicators for the discretionary layer on top
Outcome
Signals live in code instead of in my head, which makes them reviewable — and it turned into the data backbone that Frameworx was built on.
At a glance
- Cross-venue
- Arbitrage spreads compared per cycle
- 2 markets
- Polymarket and Kalshi as signal feeds
- Backbone
- Became the data layer under Frameworx
Decisions & tradeoffs
Considered: Trading on discretionary reads
Chose: Writing signal generation as code
Why: A rule in code can be reviewed, replayed, and falsified; a hunch cannot.
Tradeoff: Slower to act, and every edge case has to be handled explicitly.
Considered: Price data only
Chose: Adding prediction-market probabilities as an independent signal
Why: Prediction markets disagree with price often enough to be informative.
Tradeoff: Thin-liquidity markets need filtering or they add noise.