Mikhail Edelshtein

At Yandex, I own a multi-agent AI product end-to-end, from architecture and infrastructure to agent runtime and data integrations. I migrated it from a Python prototype to a production-grade Go implementation and helped accelerate launch by one quarter.

Previously at Twin AI, I built a fully streaming voice and chat assistant pipeline spanning speech recognition, LLMs, and text-to-speech. I reduced response latency by 2× to below 200 ms time-to-first-word while supporting session continuity through failures and pod restarts.

At MTS, I developed data access layers and observability for high-load fintech microservices serving millions of users. My work improved uptime to 99.9%, reduced API latency by 61%, and lowered database load by 39.8%.

Across these products, I work in Go on distributed microservices, streaming pipelines, databases, messaging, and cloud-native infrastructure. I enjoy taking systems from architectural decisions through production launch.