Adel Kaleche

Rust engineer focused on the seam between LLM serving and on-chain payment protocols. I build small, spec-pinned cores — wire-format-correct, benchmarked under release profile, golden-tested against the reference implementation — and the production glue around them.

Current work — x402-batch-settlement (Rust): the first Rust implementation of the x402 V2 batch-settlement scheme. A cryptographic core (EIP-712 domain, channelId, voucher digest, secp256k1 recovery) byte-exact against viem, pinned by 5 golden tests to authoritative vectors. The full economic loop — deposit → off-chain vouchers → claim → settle → recipient — is demonstrated end-to-end on Base Sepolia, single process, with on-chain receipts as the source of truth. Steady-state voucher overhead: p99 ≈ 0.39 ms (release, target-cpu=native), ~13× under the 5 ms target, with no per-payer trust window. Two upstream contributions opened: a spec-feedback PR to x402-foundation/x402 and a batch-settlement proposal to x402-rs.

Around it — the api router: an OpenAI-compatible Rust router (axum / tokio / reqwest) that gates each LLM request behind an x402 payment, forwards to an upstream provider, and reports per-stage timing in microseconds. The harness used to measure three payment modes (sync settle, async settle, batch-settlement) and document the latency / economic-risk frontier honestly.

Background spans Rust, Python, TypeScript and SQL. Comfortable owning the path from protocol spec to production-grade systems code.

Open to remote-friendly engineering roles and collaboration around payment protocols, agentic infrastructure and LLM serving.