Forward Deployed Engineer – Foundation (CFDE-F)

What the role is and why it exists. How FDEs differ from solutions engineers, consultants, and product engineers. The deployment lifecycle: post-sale handoff, discovery, scoping, build, deploy, iterate, hand-back. How FDEs are measured: deployment velocity, customer success, product feedback quality. Career paths: FDE to Senior FDE to Head of Customer Engineering to founder. The “T-shaped” profile: deep technical skills plus broad execution skills.

Stakeholder mapping

Running discovery sessions with enterprise and SME clients. Identifying the real blocker when the client cannot articulate it. Asking the right questions before writing a single line of code. Translating vague business pain into a scoped, buildable technical problem.

Architecture diagrams and decision logs

Translating scoped problems into technical requirements. Writing architecture briefs that a client’s CTO or technical lead can review and approve. System design documents. Decision logs that make reasoning visible. Making work handoff-ready.

Messy data wrangling and exploratory analysis

You never start from scratch. How to land in an unfamiliar codebase, understand it quickly, identify integration points, and work within constraints you did not design. Reading undocumented APIs. Working with tribal knowledge. Building on top of messy reality. Ad-hoc exploration using Python, Pandas, SQL, Jupyter to understand the client’s data reality.

Data ingestion, transformation, Spark, and Airflow

REST and GraphQL API design and consumption. Connecting systems that were not built to talk to each other. SQL and NoSQL database pragmatism: choosing and justifying the right tool. Vector databases. Production-grade data engineering: building reliable, scalable pipelines that feed AI systems.

Working across OpenAI, Claude, Gemini, Mistral, and open-source models (Llama, DeepSeek, Qwen). Practical differences in API design, pricing, latency, capability, and compliance posture. Building a model selection framework: when to recommend which provider to a client and justifying the choice on cost, performance, and regulatory grounds.

Prompt versioning and testing

System prompts, structured outputs, chain-of-thought, few-shot learning, guardrails. The difference between a demo prompt and a production prompt: reliability, edge cases, cost, latency. Building prompt systems that hold up when real users do unexpected things. Prompt versioning and regression testing.

Document ingestion, chunking strategies (and why chunk size matters more than most people think), embedding models, vector databases (Pinecone, Weaviate, ChromaDB), retrieval, hybrid search, re-ranking, generation. Building a complete RAG pipeline against real-world data. The single most common FDE build task in 2026.

Building evaluation suites for AI systems: faithfulness, relevancy, hallucination rate, regression detection, bias, grounding gaps. Automated evaluation vs. human evaluation. Evaluation metrics a client can understand and run themselves.

Multi-agent orchestration, human-in-the-loop, agent failure modes

Agent frameworks: LangGraph, CrewAI, OpenAI Agents SDK, MCP, tool orchestration. Single-agent and multi-agent system design. When to use agents vs. when a simpler pipeline is better. Foundation-level FDEs should know how to make two agents call each other; Professional handles the scary failure modes.

Environment management (Docker, Cloud CLIs)

CI/CD pipelines for AI applications. Monitoring, basic observability, alerting. Cost optimisation: token budgeting, caching, model routing. Deploying on Vercel, Railway, AWS Lambda, or lightweight cloud infrastructure. Model versioning and experiment tracking. Scoped to what an SME FDE actually does, not enterprise Kubernetes clusters.

When to fine-tune vs. prompt-engineer. Making the business case to a client. LoRA/QLoRA. Training data curation and quality. Evaluation gates before deploying a fine-tuned model. Version control for models. Cost implications and tradeoffs.

Handoff documentation, success metrics definition

Getting from working demo to live production. The 80% of work that happens after the demo works. Onboarding client teams to use the deployed system. Reducing resistance by tailoring to real workflows rather than forcing workflow changes. Measuring adoption. Defining success metrics with the client.

Scope management and expectation setting

Explaining technical architecture to non-technical people. Running effective client meetings. Status updates that build trust. Managing scope creep and expectation setting. Saying no constructively. Written communication: implementation guides, decision logs, handover docs.

Problem decomposition under time pressure, MVP prioritisation

The FDE’s core thinking skill. Taking a massive, ambiguous problem and breaking it into shippable chunks. Prioritisation: what to build now vs. defer. Working under time pressure without sacrificing production quality. Making and documenting assumptions. Shipping MVPs that are genuinely viable, not embarrassing. Breaking work into 1-week milestones. Delivering a “walking skeleton” early.

Realistic engagement simulation. You receive a business brief from a mock client with ambiguous requirements, a messy dataset, and a tight deadline. Deliverables: discovery notes, architecture document, working deployed prototype, client-facing documentation, and a presentation. Presented to a panel playing sceptical client stakeholders. Graded on: does the solution work, is the code handoff-ready, did you solve the right problem, can you communicate your decisions under pressure. Time-boxed to 48 hours for the build, 30 minutes for the presentation and Q&A.