Forward Deployed AI Engineer

About the Role
This is a forward-deployed, hands-on engineering role for people who’ve lived through multiple technology waves and know the difference between what’s exciting and what survives production.

You’ll embed with enterprise teams to build and harden LLM-powered systems under real constraints: hallucinations, retrieval failures, agent misbehavior, silent regressions, cost explosions, and shifting data distributions. You own systems end-to-end, from architecture through post-launch behavior.

This is not a research role, a prompt-only role, or a feature factory.
It is a delivery role for engineers who can make stochastic systems behave well enough that businesses rely on them.
‍

Examples of what You’ll Own:
Own production AI systems end-to-end

Design, build, deploy, and operate LLM-powered systems in production.
Own reliability for probabilistic systems: hallucinations, grounding, drift, latency, and cost.
Make tradeoffs explicit when correctness, speed, and cost are in tension.
LLM, RAG, and/or Agentic Systems experience

Build and debug RAG pipelines: ingestion, chunking, retrieval quality, reranking, grounding, and evaluation.
Design agent workflows that interact with tools, APIs, and data without looping, stalling, or hallucinating authority.
Identify when agents are the wrong abstraction and kill them early.
Evaluation, Observability, and Drift

Build and maintain evaluation frameworks for LLM outputs (offline, online, human-in-the-loop).
Detect silent regressions caused by model updates, prompt changes, or data drift.
Instrument systems so failures are observable, explainable, and actionable.
MLOps & Infrastructure

Operate AI systems in real cloud environments with CI/CD, monitoring, and rollback strategies.
Manage model/provider changes, versioning, and rollout strategies safely.
Control costs in systems where usage is nonlinear and failure is expensive.
Client & Stakeholder Delivery

Lead technical delivery when systems misbehave — including explaining failures to non-technical audiences.
Embed directly with client business makers, engineers, PMs, and stakeholders.
Translate ambiguity into decisions, constraints, and execution plans. ‍

Who This Is For
You’re likely a fit if you:

Have 5+ years building and operating production systems.
Have shipped and owned LLM-powered systems in production, not just prototypes.
Have dealt with hallucinations, retrieval failures, or agent misbehavior beyond prompt tweaks.
Have built or maintained LLM evaluation pipelines and know their limits.
Understand model drift, data drift, and behavioral regression in live systems.
Are strong in Python and comfortable building backend services, pipelines, and workers.
Have real experience with cloud infrastructure, CI/CD, monitoring, and incident response.
Can explain to stakeholders why AI systems fail — without hiding behind hype.
Value intellectual honesty, low ego, and responsibility when things break. ‍ You are not a fit if:
Your AI experience is mostly API integration or demos.
You expect prompt engineering alone to solve reliability.
You haven’t owned AI behavior after launch.
You want deterministic guarantees before you’re willing to take responsibility.
You avoid accountability when systems fail in production.