Talon Labs is a quantitative trading and engineering shop. We make markets on perp DEXs, carry funding across venues, and research on-chain flow at the transaction level.
Everything we trade runs on infrastructure we wrote ourselves, in Rust, on hardware we operate. It is the same infrastructure we build for other people.
Three one-tick books, quoted continuously, colocated in the venue's region.
Fair value is anchored to a multi-exchange perp consensus with event-age freshness gating. Inventory is controlled by a closed-form Avellaneda-Stoikov lean rather than a directional forecast.
Delta-neutral basis capture across venues that disagree about the price of time.
Rest the expensive leg, cross the deep one on each partial fill, hedge the delta on every wake. Every venue's funding is read from its own API in the units it actually pays.
The full swap firehose, decoded in-process and recorded to an append-only tape.
Nine AMM families decoded live: every swap, every token-balance delta, the router each trader used. Flow-quality measurement rides every row as data, never as a threshold frozen the day it was written.
Trading systems are unforgiving in a specific way: the failure modes are silent. A gate that never matched. A unit read one decimal off. A metric emitted into the void. Each one reads as healthy right up until it costs money.
EIP-712 and msgpack order signing, session-key auth, self-hosted validator nodes, and adapters that model each venue's actual physics (cancel reliability, order lifetime, liquidation source) instead of assuming they all behave alike.
gRPC account firehoses into slot-versioned in-memory state, on-chain layouts validated against landed transactions, and columnar tapes sized for research replay at gigabytes a day.
Build, sign, submit, confirm pipelines with typed failure taxonomies, cross-chain treasury routing over canonical bridges, and a credential model where a key can only do what it was explicitly granted.
Self-hosted end to end: k3s, Prometheus and Grafana, CI on our own runners, checksum-verified deploys, and liveness invariants that watch the data streams a consumer actually ingests rather than whether a process is running.
Building something that needs to touch a venue, a chain, or both.
hello@talonlabs.io