AI engineer for enterprise workflows

GeneralMind builds AI Systems of Action for complex enterprise workflows. We deploy AI employees that automate messy operational handovers across supply chains - processing orders, validating invoices, coordinating suppliers, and handling exceptions directly on top of enterprise systems like SAP, Oracle, JDE, and Dynamics. Our platform combines AI agents, operational context graphs, and real transaction data to achieve Autopilot-grade automation in real operational environments. We work closely with leading global companies across manufacturing, retail, and logistics to deploy AI into mission-critical workflows.

• Own the agent layer end-to-end: prompt and context design, tool use, orchestration, evals, and long-term technical direction.
• Be the person who understands why an agent made the decision it made - and who other engineers come to when a model behaves in ways they can’t explain.
• Set the bar on how we build with LLMs: eval-driven development, context engineering, structured outputs, cost and latency discipline.
• Make foundational decisions with conviction - which model, which pattern, which guardrail - document the reasoning, and evolve them as the system grows.
• Design agents that survive real enterprise environments: ambiguous inputs, flaky upstream systems, exception cases that weren’t in anyone’s spec.
• Build the eval harness that lets us ship confidently: offline evals, online evals, regression suites, human-in-the-loop review.
• Think in failure modes: hallucinations, silent drift, prompt injection, tool misuse. Build the detection before you need it.
• Own cost and latency as first-class product requirements, not afterthoughts.
• Design agent architectures that compose cleanly: tool selection, state, memory, planning, escalation.
• Build the context layer that makes agents actually useful on top of SAP and Oracle data - retrieval, grounding, schema awareness, auditability.
• Evolve agents without breaking live customer deployments; manage prompt and model versioning with intention.
• Know when the right answer is a bigger model, a better prompt, a fine-tune, a tool, or just classical code.
• Detect regressions in agent behavior before a customer does; build the evals and observability to catch them early.
• Own the operational posture of the AI systems you ship: alerting, runbooks, replay, on-call readiness.
• Identify structural debt in prompts, pipelines, and context before it compounds; refactor with thought.

• 6+ years building production software, with clear ownership of LLM-powered systems in real customer environments - not demos, not notebooks.
• Non-negotiable. You’ve shipped agents or LLM features that run on live traffic. You have strong opinions about prompting, context, tool use, structured outputs, and what breaks in production.
• You don’t ship LLM code without a way to measure it. You’ve built eval pipelines that actually catch regressions, not just vibes.
• You’re fluent in Python and comfortable in production service codebases. You make deliberate calls about async, framework choice, and performance.
• Hallucination containment, idempotency on tool calls, graceful degradation when a model is wrong - part of how you design, not patched in later.
• You’re comfortable being the person who holds the full agent picture - from prompt to production incident. You identify what’s broken or missing before you’re asked.
• We’re a strong team with broad coverage. What we’re hiring for is someone who brings a 10x improvement in an area we’re currently underinvested in. That might be agent evaluation, context engineering on enterprise data, agentic reliability, deep NLP/LLM internals, or domain-specific grounding. We don’t know exactly what shape it takes - but you should know what yours is, and be able to articulate why it matters.

• Условий в вакансии нет