Trace Evidence Jobs in Michigan (NOW HIRING)

We build and operate production AI voice agents that hold real phone conversations in a regulated healthcare setting, plus the machine learning and LLM pipelines around them. This is one seat that spans four disciplines that rarely come together: real‑time systems, LLM engineering, traditional machine learning, and serious cloud infrastructure, all in production, all with real consequences. If you are the kind of engineer who gets restless doing one thing, this role is the opposite problem.

• Streaming, low‑latency speech‑to‑speech systems built on modern LLMs
• Telephony and real‑time media (call control, live audio streaming)
• Audio handling and the quirks of real human conversation (interruptions, timing, noise)
• Concurrency on a latency‑sensitive path, where p99 matters and a stall is something a caller hears
• Wrapping nondeterministic models in deterministic control so they behave reliably in production
• Multi‑model pipelines, prompt design, and cost/latency budgeting
• Evaluation harnesses, including LLM‑as‑judge and automated agent‑tests‑agent approaches
• Agentic tooling that gives AI systems safe, structured access to infrastructure

• End‑to‑end ML pipelines: feature engineering, model training, and scheduled inference
• Imbalanced, messy real‑world data; calibration and explainability for non‑technical consumers
• Turning research notebooks into reproducible, auditable production pipelines

• Infrastructure as code across multiple environments (we run on AWS)
• Managed compute, data, streaming, and orchestration services
• Security engineering in a regulated setting: encryption, least‑privilege access, strict data‑handling discipline
• Observability and telemetry‑driven debugging, tracing a production issue from a metric anomaly to root cause

Occasional full‑stack work on internal tools, and an engineering workflow that leans heavily on AI coding assistants, with human accountability for every change.

• Ship and debug code on a live, real‑time voice pipeline where latency and correctness are user‑facing
• Design control systems around LLMs: guardrails, budgets, watchdogs, safe fallbacks
• Build and operate LLM evaluation and batch‑analysis pipelines
• Own traditional ML workflows from data to scheduled production inference
• Trace production issues from a metric anomaly to root cause, including building the evidence when the cause is a vendor

• 7+ years building and operating production backend systems, with strong general‑purpose programming skills (we work primarily in Python)
• Experience running distributed systems in the cloud; comfortable debugging from telemetry to root cause
• Hands‑on production experience with LLMs or generative AI (any provider or framework), plus the judgment to know when not to use a model
• Working fluency across the traditional machine learning lifecycle (you productionize; you do not need to publish)
• Disciplined in a regulated environment: small, reviewable changes and careful handling of sensitive data

• Real‑time media or telephony experience
• Front‑end / full‑stack ability
• ML pipeline experience, vector search, or embeddings
• Fluency with AI coding assistants (our workflows assume them, with human accountability for every change)

Smallest correct change wins. Every behavior change is validated against the live system. Evidence over opinion in debugging. Code review is rigorous. Safety and privacy gate everything.

This role is open only to US citizens and lawful permanent residents (Green Card holders). We cannot consider candidates who require visa sponsorship now or in the future, and we are unable to make exceptions of any kind.

• Your LinkedIn profile URL
• A phone number where we can reach you

A resume is welcome but optional; the two items above are required.