Forward Deployed Engineer - Artificial Intelligence/Machine Learning

Forward Deployed Engineer (FDE)

A client-facing engineering role delivering AI-driven solutions directly with clients. This is not a backend seat.

Key Responsibilities:

• Client-facing ownership: Presents to clients, runs discovery conversations, translates business problems into technical solutions, and owns delivery end to end.
• Production LLM systems: Has designed and shipped LLM-powered systems in production. ML theory or POC-only experience is not enough.
• RAG and agentic workflows: Architects RAG pipelines end to end and builds agentic workflows with tool use and multi-step orchestration.
• Evals and model decisions: Writes and runs evals to measure quality. Makes deliberate choices on model selection, routing, context management, and cost/latency tradeoffs.
• LLM failure modes: Understands hallucination, prompt injection, and degraded retrieval, and engineers around them.
• Independent full-stack delivery: Ships a working product alone. Goes from a whiteboard conversation to a working POC in days, then hardens it for production.

Tech Stack:

• Backend: Strong in one or more of Java, Golang, Python, Node.js, or .NET. Designs clean REST APIs, models data sensibly, handles auth, and reasons about performance and error handling in production.
• Frontend: Builds functional, presentable UIs in React, Angular, Vue.js, or similar.
• Databases: Fluent in SQL (Postgres or MySQL) and at least one NoSQL store. Vector DB experience (pgvector, Pinecone, Qdrant) is a strong plus.
• Knowledge bases: Experience building retrieval knowledge bases (chunking, embedding, indexing) and knowledge graphs (Neo4j, graph-based RAG).
• Cloud and containers: Hands-on deployment on at least one of AWS, Azure, or GCP. Docker is a must; Kubernetes is a plus.
• CI/CD and observability: Sets up pipelines, manages environments and secrets, and maintains logging and monitoring without a dedicated DevOps engineer.
• Security: Applies sensible defaults for auth, API keys, data handling, and client data isolation. Implements prompt injection defenses and guardrails.
• Ambiguity tolerance: Thrives with loosely defined problems, shifting requirements, and minimal hand-holding.