Technology

Sr AI Engineer

SATCON IncUnited States🇺🇸United StatesPosted Oct 22, 2026

Why This Role Stands Out

This high-impact role at SATCON Inc offers you the chance to shape cutting-edge AI systems while enjoying the flexibility of a hybrid work environment. You will thrive here if you are a data-driven engineer who excels at building robust evaluation harnesses and solving complex statistical challenges at scale. Apply today to advance your career by tackling sophisticated technical problems within a forward-thinking team.

Quick Overview

Seniority

Mid Senior

Work mode

Hybrid

Location

United States

Posted

22 hours ago

GDPRHIPAALLMPython

Job Description

Role: Sr AI Engineer

Duration: 1 year Contract

REQUIRED

Evaluation harness ownership. Has personally built an evaluation harness for a non-deterministic system and used it to make a consequential decision shipping, killing, or choosing between approaches on evidence. Candidates who cannot evidence this are not shortlisted.

Measurement and statistics. Hypothesis testing, confidence intervals, sample sizing, variance under non-determinism; field-level precision, recall and F1 versus end-to-end accuracy; calibration; and the difference between a metric that moved and a system that improved.
Dataset and error-analysis discipline. Sampling and stratification, annotation protocol, inter-annotator agreement, dataset versioning, label-budget efficiency, and finding the subpopulation a healthy headline metric conceals.
Production extraction and retrieval. Both at scale, as described above.
Expert Python 3.11+. Asyncio, enforced type annotations, Pydantic v2 modeling, packaging, profiling. Minimum five years in production, including services the candidate operated.
Event-driven services and safe change. Idempotency, deduplication, dead-letter handling, replay and backfill; schema evolution, API versioning, coordinated rollout and rollback.
Regulated personal data. Classification, minimization, redaction, retention, and least-privilege identity with managed secrets.

PREFERRED

Agentic orchestration in production, and model economics token accounting, caching, and a defensible cost per processed unit.
Online experimentation: shadow deployment, A/B and interleaved comparison, guardrail metrics, and reviewer corrections as a ground-truth stream.
Drift detection, classical ML where it beats an LLM, and a considered position on LLM-as-judge.
Fine-tuning, adapter training, distillation, or systematic prompt optimization.
Analytical fluency at warehouse scale, and the discipline to move analysis out of notebooks into reproducible pipelines.
Regulated delivery under FERPA, HIPAA, GDPR, PCI-DSS, or SOX.