Remote Applied AI Engineer — Build Scalable Agent Platform

Topsail; United States; À distance; USD 140 000 - 210 000; Plein temps

Avantages offerts par ce poste

Remote-first

Meaningful equity

Above-market compensation

AI tooling budget

Résumé du poste

Topsail is hiring an applied AI engineer to design, build, and maintain agent runtimes, tool integrations, evals, and observability that power client deployments. You will ship capabilities across deployments, not just demos, and help scale patterns from one client to the next in a week, not a quarter.

You will work with forward-deployed engineers and strategists, turning field learnings into reusable infrastructure while remaining non-client-facing by default.

Qualifications

• Must have shipped an agent or LLM system used by real users.
• Strong opinions on agent architecture: context management, retrieval, and tool design.
• Solid software engineering fundamentals: backend, APIs, reliability, and cost efficiency.

Responsabilités

• Design and build agent systems, including planning, tool use loops, memory, guardrails, and human handoffs.
• Own the harness: evals, traces, and judge-in-the-loop patterns to benchmark performance.
• Build the shared platform: MCP servers, connectors to ERPs/CRMs, Salesforce, Excel, Slack, etc.
• Keep the system model-agnostic to allow move between major providers without rebuilds.
• Pair with forward-deployed engineers to turn one-off fixes into reusable infrastructure.

Connaissances

Shipped agent/LLM system

Backend software engineer

Model-agnostic deployments

Observability & runtime

Outils

Salesforce API

Slack integrations

Description du poste

Topsail is hiring an applied AI engineer to design, build, and maintain agent runtimes, tool integrations, evals, and observability that power client deployments. You will ship capabilities across deployments, not just demos, and help scale patterns from one client to the next in a week, not a quarter.

You will work with forward-deployed engineers and strategists, turning field learnings into reusable infrastructure while remaining non-client-facing by default.