Forward Deployed AI Technical Lead

Svitla Systems Inc. is looking for a Forward Deployed AI Technical Lead to join our Forward Deployed AI Engineering Pod, a small, senior team that works directly inside a client’s environment to take AI workflows into production, one workflow at a time.

As the Technical Lead, you’ll be the technical owner of everything the Pod ships. You’ll design the architecture of each workflow, write the first code and keep writing it, lead a rotating bench of specialists (agent and MCP engineering, data, integration, document AI, evaluation, security, and more), and produce the evidence that each increment is truly done. This is one of the Pod’s three core roles that stay with the client for the entire engagement, so you’ll carry what you learn from one Pod to the next and help shape Svitla’s reusable AI engineering toolkit.

Location: USA

Employment Type: Full-time

Requirements

• Experience designing and personally building production AI or ML systems that are still running, with the ability to walk through the architecture, trade-offs, and lessons learned.
• Hands-on experience with agentic systems in production, including orchestration frameworks (e.g., LangGraph or LangChain), tool and function calling, MCP, and an evaluation harness used to validate systems before cutover.
• Experience leading a small engineering team inside a client’s or another team’s environment, setting direction while still shipping code.
• Strong Python skills and solid API and systems design.
• Experience with at least one of AWS, Azure, or GCP, and with a modern data platform such as Databricks, Snowflake, or BigQuery.
• Knowing when not to use a generative model: experience selecting and integrating traditional models (forecasting, anomaly detection, classification, clustering) as tools.
• Practical MLOps experience: reproducible pipelines, model and prompt versioning, observability, and cost visibility.
• Clear technical writing skills: design notes, delivery evidence, and runbooks that client engineers can follow.
• Confidence working directly with clients: explaining results to sponsors, handling tough questions from client architects, and disagreeing constructively.

Will be a plus

• Prior forward deployed, embedded, or consulting engineering experience.
• Depth in document AI, knowledge graphs, time series, or anomaly detection.
• Experience with commerce platforms, data catalogs (e.g., Unity Catalog), ERP or EDI systems, or BI platforms.
• OpenAI or Anthropic partner certification, or willingness to complete one in the first three months.

Responsibilities

• Own the architecture of each workflow on the client’s existing stack, including core systems, data lake or warehouse, semantic layer, governed MCP surface, and agents, and document design decisions and their reasoning.
• Write the first code and keep building: orchestration, tool contracts, evaluation harness, and integrations.
• Lead the Build phase: evaluate competing approaches on holdout data, select the right production model (frontier LLM, smaller model, or traditional model), and keep every component swappable.
• Run the Prove phase: shadow mode against live outcomes, labeled test sets, precision and recall or equivalent business measures, and the evidence the client needs to decide on cutover.
• Deliver complete Definition of Done evidence for every increment: working capability, tests, evaluation results, documentation, runbooks, and reusable-component tagging.
• Lead the specialist bench: define monthly deliverables, pair with specialists inside the client’s environment, review their work, and own the result.
• Work alongside the client’s engineers so they can maintain and extend each workflow after handover.
• Uphold security and governance together with the Cloud and Platform Engineer: least-privilege access, agents as identities, no client data leaving the client’s environment, and audit logging from day one.
• Contribute reusable components back to Svitla’s library, including connectors, harness patterns, agent scaffolding, and deployment templates, and reuse them on future Pods.