Capture the market as it happened
High-cadence order-book tapes preserve price, depth, and time-to-resolution across short-horizon event markets. The raw record stays immutable; every derived dataset is traceable back to source.
PRIVATE RESEARCH PREVIEW
Research infrastructure for markets that resolve in minutes.
SkildAI turns high-frequency event-market data into evidence-grade research—captured live, normalized across venues, and tested with controls that separate a real signal from a statistical mirage.
Built for researchers, market operators, and teams that need an auditable answer—not another backtest.
BTC / 15MRESOLVES IN 04:18
0.54ASK+1.2%
14:2814:3014:32
VENUES02aligned
CADENCE70msmedian
LINEAGEPASSverified
31.4Morder-book snapshots analyzed
805 GB+compressed research corpus
6crypto underlyings
3contract horizons
THE PLATFORM
From raw tape to a claim you can defend.
One evidence chain connects live market capture, repeatable analysis, and deployment controls.
Compare venues on one clock
A common event model aligns economically equivalent contracts across venues with different schemas, fees, and settlement rules—so differences can be measured instead of hand-waved.
Test signals without fooling yourself
Cluster-aware inference, pre-registered hypotheses, and burned holdouts prevent thousands of correlated quotes from masquerading as thousands of independent outcomes.
AWS-NATIVE FOUNDATION
Designed around the data, not around a dashboard.
Continuous recorders run close to market endpoints. Immutable objects land in Amazon S3. Interruption-tolerant Graviton workers bring compute to the corpus, cutting transfer cost and making full re-runs routine.
Amazon S3EC2 GravitonEC2 SpotEBS gp3CloudWatch
01
Capturemulti-region recorders
LIVE02
Preserveimmutable S3 corpus
805 GB+03
Computeephemeral in-region workers
SPOT04
Verifylineage + holdout gates
SEALED RESEARCH INTEGRITY
The unit of evidence is the settlement—not the quote.
Consecutive quote-seconds inside one contract all resolve to the same outcome. Treating them as independent inflated naive sample sizes by 124× to 1,042× in our first census. Our workflow corrects that by construction.
Pre-register
State the claim, metric, and rejection rule before looking at the evaluation span.
Burn on look
A data window is marked consumed before its results are exposed to a researcher or agent.
Replicate
A candidate must survive a distinct time span, venue, or implementation before promotion.
Fail closed
Missing evidence, stale telemetry, or an incomplete safety gate stops a deployment.
EARLY FINDING · PRICING EFFICIENCY
0 / 138adequately powered price × time cells beat the displayed ask plus fees.
A negative result is still a result when the methodology makes it reproducible. Aggregate findings and the effective-sample-size reference implementation are being prepared for release.
SPEAKUPAI INC. · INCORPORATED 2024
Built by people who would rather disprove a signal than deploy a fragile one.
Nicholas XuFounder
SkildAI is a research initiative of SpeakUpAI Inc. We build practical data systems at the intersection of AI-assisted research, market microstructure, and cloud infrastructure. The platform is currently in a private research preview.
WORK WITH US