Backend/infrastructure engineer with ~1 year of experience, currently MTS at a YC-backed startup building sandbox and workflow infrastructure for AI agents.
Recent production work:
- Cut persistent workspace setup from ~70s → ~10s and sandbox recovery from ~486s → ~14s while preserving sessions across restarts/code updates.
- Made workflow registration idempotent across retries, duplicate hooks and out-of-order deployments, reducing workflow-sync failures from ~90% → ~1%.
- Built reconnectable multi-client streaming with one canonical agent run per task.
- Previously built a Go backend handling 10k+ concurrent WebSocket connections at sub-100ms latency.
Systems projects:
- lsm-tree-go — LSM-tree storage engine with mmap WAL, bloom filters, sparse indexing and leveled compaction; profiled from ~250k to ~2M writes/sec and 5x BoltDB throughput on random writes.
- vase — Go pub/sub broker using a lock-free ring buffer + mmap WAL; ~62x Redis throughput at 100 subscribers in my benchmark, with zero message loss across 5M messages at 1,000 subscribers.
- Engram Engine — durable agent-memory layer using event sourcing + WASM isolation with sub-1ms sandbox overhead.
Looking for backend, infrastructure, platform, distributed-systems, storage-engine, or systems-programming roles. Especially interested in small/high-ownership teams working on runtimes, execution infrastructure, databases, developer infrastructure or reliability.
Remote: Yes
Willing to relocate: Yes, including internationally with sponsorship
Technologies: Go, distributed systems, Linux, storage engines, concurrency, mmap, io_uring, TCP/WebSockets, PostgreSQL, Redis, Kafka, Docker, gRPC
Résumé/CV: https://drive.google.com/file/d/11ASup-0rcQivWNIMR6mOoYqEkhe...
Email: ashishrathour1102@gmail.com
GitHub: https://github.com/AasheeshLikePanner
Backend/infrastructure engineer with ~1 year of experience, currently MTS at a YC-backed startup building sandbox and workflow infrastructure for AI agents.
Recent production work: - Cut persistent workspace setup from ~70s → ~10s and sandbox recovery from ~486s → ~14s while preserving sessions across restarts/code updates. - Made workflow registration idempotent across retries, duplicate hooks and out-of-order deployments, reducing workflow-sync failures from ~90% → ~1%. - Built reconnectable multi-client streaming with one canonical agent run per task. - Previously built a Go backend handling 10k+ concurrent WebSocket connections at sub-100ms latency.
Systems projects: - lsm-tree-go — LSM-tree storage engine with mmap WAL, bloom filters, sparse indexing and leveled compaction; profiled from ~250k to ~2M writes/sec and 5x BoltDB throughput on random writes. - vase — Go pub/sub broker using a lock-free ring buffer + mmap WAL; ~62x Redis throughput at 100 subscribers in my benchmark, with zero message loss across 5M messages at 1,000 subscribers. - Engram Engine — durable agent-memory layer using event sourcing + WASM isolation with sub-1ms sandbox overhead.
I write about the engineering behind these projects: https://medium.com/@ashishrathour1102
Looking for backend, infrastructure, platform, distributed-systems, storage-engine, or systems-programming roles. Especially interested in small/high-ownership teams working on runtimes, execution infrastructure, databases, developer infrastructure or reliability.