Customer Support Engineer

About Blitzy

Blitzy is a Cambridge, MA based AI software development platform on a mission to revolutionize the software development life cycle by autonomously building custom software to unlock the next industrial revolution. We're transforming how enterprises build software, turning enterprise requirements into production-ready code with an agentic software development platform that can autonomously execute 80% of the quantum of software development work.

We're backed by multiple tier-1 investors, and have proven success as founders of previous start-ups.

Location:

Cambridge, MA (On-site)

Compensation: $100,000 - $140,000 salary + equity

The role is to support our clients and ensure a stable environment across the full lifecycle: installation, ongoing upgrades, and day-to-day operation. The L2 Support Engineer works alongside L1 to triage and resolve issues, and escalates unresolved defects to engineering. It operates across Kubernetes, Docker, and the major cloud providers.

• Customers' issues are resolved faster and escalated cleaner.
• Recurring problems turn into runbooks, dashboards, and alerts, not repeat tickets.
• Engineering trusts your escalations because they come with proof, not guesses.
• Customers trust your communication because it's clear, honest, and on time.

• Deploy and install the platform into customer environments, and troubleshoot installation issues.
• Support ongoing upgrades and day-to-day operation, keeping customer environments stable.
• Work alongside L1 to triage and resolve customer-reported issues, driving them to resolution or escalation.
• Diagnose failures across the stack: compute, networking, storage, and the services running on it.
• Reproduce issues safely against live (often multi-tenant) environments using read-only diagnostics first.
• Build and maintain dashboards, monitors, and runbooks so recurring issues get faster to fix: or stop recurring.
• Write up clear, evidence-backed escalations and post-incident notes.
• Communicate status and resolution to customers clearly and on time.

• Distributed-systems debugging. Reason about a request crossing multiple services, queues, and network hops, and isolate which hop failed. You debug by forming a hypothesis and confirming it with evidence (logs, pod state, queue depth, DB rows), not by guessing.
• Kubernetes & Docker.
• Major cloud providers: GCP, AWS, and Azure. Hands-on with at least one deeply and able to work across the others: managed Kubernetes (GKE/AKS), cloud logging, IAM/auth basics, and cloud disk/storage behavior.
• Strong monitoring & observability practice. Fluent with an APM/observability stack (Datadog or equivalent): log queries, correlating across services by request/trace IDs, reading traces, and building dashboards and alerts. You reach for the data before theorizing.

• Python and Redis literacy.
• Basic message queueing. Command transport runs over a message queue (Redis/rq). Comfort inspecting queue depth, backlogs, and stuck/failed jobs; concepts transfer from any broker.
• Networking & Web Sockets. Many of our hardest issues are connection problems:
  Web Socket/Socket.
  
  IO drops, NAT/idle/LB timeouts, half-open sockets, DNS-vs-routing, TLS. Tell a transport fault from an application fault.
• SQL / PostgreSQL. Query operational tables to confirm what the system recorded.
• Source-control platforms. Git Hub (incl. Git Hub Enterprise Server), Azure Dev Ops, and/or Git Lab, clone/push/pull, access tokens, app credentials, and their failure modes.
• CI/CD, Helm & deploy integrity. Many “sudden regressions” are a bad or partial deploy: check what version is actually running before chasing architecture theories. Helm and container deploy pipelines expected. ArgoCD is a plus.
• Secrets management. Comfort handling secrets, credentials, and certificates safely, ideally with Vault (strongly preferred).
• Linux and Windows. Workloads run on both; comfort triaging on each OS (process inspection, file system, basic networking).
• Methodical, evidence-first temperament. Hold several candidate causes at once, run the cheapest disconfirming check…