Stefanini is looking for an AI Engineer, Dearborn, MI (Onsite)
For quick apply, please reach out Adil Khan at 248-728-6424/adil.khan@stefanini.com
You will be responsible for developing intelligent programs, cognitive applications and algorithms for data analysis and automation, leveraging various AI techniques such as deep learning, generative AI, natural language processing, image processing, cognitive automation, intelligent process automation, reinforcement learning, virtual assistants and specialized programming.
Responsibilities
Understand business requirements and develop AI algorithms, models and programs to solve complex problems, generate recommendations, extract patterns, make predictions, interpret sensor data (images, sound), orchestrate automation and enable self-service capabilities.
Perform large-scale experimentation and develop data driven applications that translate data into actionable intelligence.
Drive innovative applications of Artificial Intelligence tools and techniques such as deep learning, generative AI, natural language processing, image processing, cognitive automation, intelligent process automation, reinforcement learning, virtual assistants and specialized programming.
Research and optimize AI technologies to enhance efficiency and accuracy of data analysis and create more efficient automation.
Architect and deploy the production multi-agent orchestration layer (interpreter/orchestrator, NL-to-SQL agent, visualization agent, RCA/RAG agent, report composition agent, notification agent), using modern agent frameworks with state management and checkpointing rather than ad-hoc loops.
Design and productionize RAG pipelines (chunking, embeddings, hybrid retrieval, reranking) grounded in approved schemas, engineering documentation, and historical issue records.
Own BigQuery integration and enforce safe, least-privilege, validated execution of LLM-generated SQL.
Build CI/CD, containerization, and infrastructure-as-code for deploying agent services on GCP (Cloud Run/GKE, Vertex AI).
Implement evaluation pipelines and observability/tracing for every agent (golden datasets, LLM-as-judge scoring, regression alerts) so quality is measurable, not assumed.
Implement guardrails, prompt-injection defenses, and human-in-the-loop approval checkpoints to ensure correctness and safety before any output triggers downstream action.
Design cost/latency optimization strategies, including tiered model routing (cheap filter models vs. high-capability deep-dive models) and caching.
Integrate validated outputs with operational systems (Salesforce ticketing, driver/site-manager notifications) and report export pipelines (PDF/HTML/spreadsheet).
Collaborate with data scientists to productionize prototypes (anomaly detection, diagnostic agents) into scalable, monitored services.
Establish versioning, testing, and safe rollout practices (canary/shadow deployments) for evolving agent logic.