Srikar Yaganti

I'm Srikar — a distributed systems engineer. Most of my career has been event-driven architectures, real-time data pipelines and multi-chain infrastructure at scale. The last stretch has been a deliberate focus on the high-performance end of that: single-digit-millisecond services, and the measurement discipline to prove a system is actually as fast as it's claimed to be. Mostly Go and Rust these days, and I'll own the whole path from bare metal to the product surface.

What I work on

Distributed systems. Event-driven architectures and the failure modes that come with them: idempotency and replay, backpressure, partial failure, and exactly-once claims that turn out not to be. Multi-chain infrastructure across 15+ EVM networks, data pipelines in the millions of records a day, and service topology that stays debuggable as the service count grows.

Low-latency engineering — the recent focus. Hot-path work where the wins come from data layout rather than cleverness: fixed-point integers instead of heap-allocated decimals, dense indexes instead of trees when the key domain is bounded, incremental derivation instead of full rebuilds, and allocation removed until the garbage collector stops mattering. Order books and matching engines are where I've applied most of this lately, but none of the techniques are specific to them.

Real-time data. Multi-source WebSocket ingestion and normalization. Record-and-replay feeds that are wire-compatible with the original, so a consumer points its existing client at the recording and can't tell the difference. Staleness detection that separates "everything is old" from "one feed went quiet while the other kept moving" — the second one is what actually costs you.

Measurement. Reproducible benchmark harnesses with the raw samples published, not just the percentiles: four message brokers against two serialization formats, eight JSON encoders, pgx/sqlc against GORM. Pinned image digests, caveats stated before the numbers, and a public retraction when the data doesn't support what I said it would.

Platform and reliability. Observability architecture — scrape topology, service discovery, cardinality and cost. Load-testing infrastructure for signature-gated APIs, where you have to manufacture a funded population before you can send request one. And the unglamorous cost work: logging was the second-largest line item on the bill until it wasn't. Security-first, SOC2 and SLSA Level 3.

Applied cryptography. EIP-712 and ERC-7739 nested typed-data signing, contract-wallet (ERC-1271) signature flows, gasless order management on Polygon. I validate ported implementations byte-for-byte against reference vectors — which is how I found open bugs in two official exchange SDKs.

Background

Head of Engineering at a Web3 treasury platform scaling to 200+ DAOs, where I owned the architecture for multi-chain infrastructure across 15+ EVM blockchains — and stayed hands-on in the code the whole way. Before that, backend engineer at a fintech unicorn, holding services in the single-digit-millisecond range at 3M+ daily requests and running data pipelines over 2M+ records a day. Infrastructure and data platform engineering at Amazon.

Recent side work: a prediction-market trading platform built end-to-end — per-market order book service, automated market making across 1,000+ markets with vault risk management, and real-time sports data ingestion; an algorithmic trading system for prediction markets (Go decision engine, real-time Chainlink/Binance/Polymarket ingestion, TypeScript execution service, multi-gate risk system that fails closed); a cross-venue arbitrage engine that taught me considerably more by failing than by working; an air-gapped RAG platform for defense and aerospace document intelligence; and a bare-metal crypto market data API.

Outside work: dad to 2 dogs, riding motorcycles, reading anthropology and psychology, keeping up with interesting science.

Writing

Measurements, and the occasional postmortem on something that didn't work. Every benchmark post ships the command that produced it and the raw data behind each number.