Blockchain2024
GMXSUBSTREAMS
Sub-100ms TimescaleDB queries on GMX protocol events. 87% faster deploys via Rust CI/CD engine.
<100ms
PnL / horizon queries
87%
Faster deploys (Rust CI/CD)
60%
Hot-path IO reduction
SCROLL
Sub-100ms TimescaleDB queries on GMX protocol events. 87% faster deploys via Rust CI/CD engine.
GMX protocol indexing for a user-facing trading dashboard with PnL and Horizon Analysis at production query speeds.
Substreams pipeline in Rust with TypeScript DB sink, TimescaleDB hypertable partitioning, and a Rust/Axum engine coordinating 5 GCP services.
Index GMX Substreams into TimescaleDB with sub-100ms responses, plus a CI/CD orchestration engine cutting deploy time 87%.
Rust module consumes GMX vault, position, and swap events; TypeScript sink batches writes into Timescale.
Hypertables by time and market; continuous aggregates for horizon curves; retention policies for raw vs rollup tiers.
Expose pre-aggregated endpoints aligned to the stealth desk’s PnL and leaderboard widgets.
Axum/Tokio service orchestrates GCP APIs for build, deploy, and task fan-out with auditable run logs.
Health checks, rollback hooks, and artifact pinning so reorgs or RPC drift could be replayed safely.
PnL and Horizon Analysis charts stayed interactive on multi-terabyte Timescale datasets because continuous aggregates and tuned chunk intervals replaced full scans for every dashboard interaction.
PnL and Horizon Analysis charts stayed interactive on multi-terabyte Timescale datasets because continuous aggregates and tuned chunk intervals replaced full scans for every dashboard interaction.
Rust/Axum CI/CD engine coordinating Cloud Build, Artifact Registry, Cloud Run, and Cloud Tasks cut release cycles from roughly two hours to fifteen minutes — fewer manual handoffs between consoles.
Hypertable partitioning plus compression policies reduced hot-path IO enough that traders saw consistent sub-second rollups even during volatile funding windows.
Native time-series chunking and continuous aggregates matched trading-window queries without bolting on a second OLAP store.
Compression policies paid for themselves once raw swap rows crossed the TB mark.
Strong typing and async IO for coordinating five GCP services from one binary with fewer shell scripts.
Cut deploy time 87% by replacing click-ops with idempotent Cloud Tasks workflows.
We take on a small number of projects at a time. If the problem is hard, we're interested.