The roadmap to becoming an FDE
8 stages. Follow them in order, or start with the one where you are weakest.
Foundations
Understand the role and how an FDE creates value.
Skills: The FDE role · How it differs from SWE and solutions engineering
- GuidesMessy client Excel files: count the missing cells before pandas adds them up
- GuidesFDE interview prep: enough DSA to pass, then focus on decomposition
- GuidesWhy, what, how: split an FDE engagement into three questions before writing code
- GuidesFDE or Solutions Architect: the dividing line is who commits to production
- GuidesWhy AI labs are hunting for Forward Deployed Engineers
- Books & coursesHarvard's CS50 SQL: seven weeks on table design before you touch customer data
- Books & coursesAndrew Ng's Machine Learning Specialization: three introductory courses, read through an FDE's eyes
- Books & coursesReading The Pragmatic Programmer as an FDE: three habits for working alone at a client site
Broad engineering
Handle everything from data and APIs to infrastructure yourself.
Skills: Python · SQL · APIs · Docker
- GuidesISO 20022 for FDEs: check the payment data before bringing AI into a bank
- GuidesMulti-tenancy for AI systems: stopping cross-customer data leaks at four layers
- GuidesHL7 v2, FHIR and Epic: integrating AI with hospital data through two pipelines
- GuidesInheriting a legacy Hadoop cluster at a bank or telco: read HDFS, YARN, MapReduce and HBase before you touch it
- GuidesSettle the grain before you draw a table: building a star schema beside a client's transactional database
- GuidesAdding AI to a client's existing event queue without rewriting the system
- GuidesLatency, throughput, scalability: answering the client's infrastructure team with measurements
- GuidesKafka and Flink for FDEs: wiring a customer's data stream into a real-time scoring model
- GuidesMLOps for FDEs: five things to build before a model runs in a client's system
- GuidesThe demo works at your desk but fails on the customer's network: debugging DNS, proxies, CORS and cookies
- GuidesGit in a client's repo: identity, remotes, credentials and review rules
- GuidesReading a customer's Java, Go or Node.js codebase when you only know Python
- GuidesWhat FDE recruiters look for: code gets you in, customer communication sets the ranking
- GuidesData lake, warehouse or lakehouse: read the client's data architecture before writing your first ML pipeline
- ToolsMetabase: self-service dashboards on a client's database, built in an afternoon
- ToolsRetool for FDEs: build internal tools in hours, and know when to go back to React
- Books & coursesAWS Cloud Practitioner Essentials: 13 modules that teach you to talk to a client's IT team
- Books & coursesWorking Effectively with Legacy Code: the 2004 book that helps FDEs add AI features to untested code
- Books & coursesKleppmann and Riccomini update DDIA: storage is moving to object stores, and the trade-offs are moving too
- Books & coursesThe Data Warehouse Toolkit: Kimball's book on keeping dashboards from rewriting past numbers
- Books & coursesFundamentals of Data Engineering, four years on: the map to take on customer sites
- Books & coursesData Engineering Zoomcamp: a free course for aspiring FDEs who need to build data pipelines from scratch
Applied AI
Use LLMs and agents to solve real problems.
Skills: LLMs · Agents · Model evaluation
- GuidesSub-agent, skill or MCP server: where a client's workflow belongs
- GuidesExtracting invoices and contracts with LLMs: schemas, business-rule checks and source citations
- GuidesA2A and MCP: when your agent has to hand work to another vendor's agent inside a client's system
- GuidesHands-on: selecting, balancing and ordering few-shot examples from real client data
- GuidesWhen the customer's data is a network: Neo4j, Neptune and GraphRAG
- GuidesChunking client documents: measure before you cut, then add context to every chunk
- GuidesClassical ML or deep learning: choose by the shape of the client's data, not by fashion
- GuidesRead the model card before choosing Qwen, Llama, Gemma or Mistral for a Vietnamese project
- GuidesCalling the Claude and OpenAI APIs directly: write your own tool-use loop, then compare with Gemini
- GuidesZero-shot, few-shot, CoT or ReAct: test all four on one ticket set to learn when a heavier technique is needed
- GuidesHybrid search and reranking: building a RAG pipeline that finds the right contract number or product code
- GuidesChoosing an embedding model for Vietnamese data: build a 50-question test set before trusting the leaderboard
- GuidesContext engineering: choosing which customer data the model gets to see
- GuidesLLM basics for FDEs: tokens, context windows, temperature and why models make things up
- GuidesAfter go-live, an agent's real eval set lives in customer traces
- GuidesFine-tuning or RAG: diagnose the problem before you pick the technique
- ToolsPydantic AI: typed tools and evals let FDEs switch models without gambling
- Toolsn8n for FDEs: build an LLM workflow in an afternoon, but know when to move to code
- ToolsLangGraph and interrupt: making an agent stop and ask before it does something that matters
- Books & coursesHugging Face's MCP Course: FDEs should study Unit 3 closely and move quickly through Unit 2
- Books & coursesBerkeley RDI's Agentic AI course: what to study before you take agents into the enterprise
- Books & coursesHugging Face's LLM Course: how the first three chapters take an FDE from pipeline() to fine-tuning
- Books & coursesNeural Networks: Zero to Hero: Andrej Karpathy's eight lectures, from 150 lines of code to GPT
- Books & coursesPractical Deep Learning for Coders: nine lessons from code to a working model
- Books & coursesBuilding with the Claude API: Anthropic's free course covers half the engineering of the FDE job
- Books & coursesLLM Zoomcamp: a free RAG course that also teaches evaluation and monitoring
- Books & coursesGenerative AI with LLMs: the DeepLearning.AI and AWS course that walks you through the full lifecycle of an LLM application
- Books & coursesAndrew Ng's Agentic AI course teaches four design patterns, but its best lesson is evals
- Books & coursesReading "Designing Machine Learning Systems" as an FDE: start with chapters 3, 4, 9 and 10
- Books & coursesChip Huyen's AI Engineering: a book with little code that teaches FDEs how to decide
- Books & coursesHugging Face's free AI Agents course: the 30% pass mark is the lesson worth learning
Customers
Find the real problem and communicate clearly.
Skills: Customer discovery · Writing proposals
- GuidesDPAs, data terms and procurement: what FDEs need to know so projects don't stall in legal
- GuidesSaying “no” to a bad design without losing the client
- GuidesHow to agree with the client on what ‘success’ means before you write any AI code
- GuidesWhen the shift supervisor doesn't trust AI: how an FDE overcomes resistance in a pilot
- GuidesTen minutes with the client's leadership: turning pilot results into a signed decision
- GuidesWhen the project sponsor quits mid-deployment: how an FDE keeps the work alive
- GuidesWhy was this loan rejected? Use SHAP to explain it and LIME to cross-check
- GuidesOne deployment, four audiences: how FDEs work with a client's engineers, business teams and executives
- GuidesThe customer says the dashboard is slow, but what they lack is conversion: a discovery lesson for FDEs
- Books & coursesThe Pyramid Principle: the book that teaches FDEs to write so client executives get it from the first sentence
- Books & coursesPeter Block's Flawless Consulting: the book for FDEs whose solution works but goes unused
- Books & coursesThe Culture Map: hear American, German and Japanese client feedback correctly before it wrecks a deployment
- Books & coursesNever Split the Difference: what a former FBI negotiator can teach FDEs about holding scope
- Books & coursesCrucial Conversations: the 2002 book that teaches FDEs how to talk when a meeting is about to fail
- Books & coursesRob Fitzpatrick's The Mom Test: a handbook for FDEs on questioning customers when everyone lies
Deployment
Put the solution into production with real data.
Skills: Data integration · Monitoring · Incident handling
- GuidesBuild a webhook receiver that withstands forged signatures, replays and duplicate events
- GuidesEncryption, tokenization or masking: choosing the right tool for each customer data field
- GuidesL4 and L7 load balancers and reverse proxies: running your service behind a customer's network
- GuidesCutting LLM latency and cost: four levers and the order to pull them
- GuidesDesign error responses with RFC 9457 so customer ops teams can fix incidents themselves
- GuidesEdge AI in factories and shops: choosing between TensorRT, LiteRT and ExecuTorch
- GuidesDVC in practice: make every new client data drop a traceable commit
- GuidesBuild CI/CD for ML models with CML: post metric comparisons on every pull request
- GuidesHands-on: build a Slack approval gate before your agent issues refunds or sends emails
- GuidesWhen an agent must stop: four guardrails and the moment to hand over to a human
- GuidesGuardrails for a client's LLM app: filter the input, lock down the output, add moderation
- GuidesDebugging without access to a customer's production: finding faults with logs, data samples and reproductions
- ToolsSentry for FDEs: fixing bugs on a client site you are not at
- ToolsHelm for FDEs: package a deployment as a chart and install it for many customers without touching the templates
- ToolsAWS Glue, Azure Data Factory and Google Dataflow: how to read the ETL pipeline a client already runs
- ToolsMLflow: answering “which model is running?” with runs, versions and aliases
- Books & coursesMichael Nygard's Release It!: the bedside book for FDEs who live in a client's production
- AnalysisKubeflow or SageMaker, Vertex AI, Azure ML: ask who will run the pipeline before comparing features
- AnalysisKeep logs for 12 months, delete data without delay: an API architecture problem
- AnalysisLangChain, LlamaIndex, Haystack or direct API calls: choose a framework for the client's problem, not for its popularity
Measurement
Prove value with the numbers customers care about.
Skills: Business metrics · Reporting
- GuidesReconciliation: only say "it works" once your numbers match the client's books
- GuidesWriting your first eval with Inspect AI: dataset, solver, scorer and reading the log
- GuidesFrom thumbs-down to eval set: turning user complaints into test cases
- GuidesWhen an LLM should say "I'm not sure": self-evaluation, confidence and the threshold for handing off to a human
- GuidesData lineage: tracing a wrong prediction back, step by step
- GuidesWhen a model is wrong but reports no errors: build your own data drift detection with Prometheus and Grafana
- GuidesHow to pick a classification threshold by the cost of each error, not the 0.5 default
- GuidesLLM-as-judge in practice: calibrating a judge against the client's expert pass/fail labels
- GuidesMeasuring ROI for customers: count work that meets the bar, subtract checking and rework
- ToolsLiteLLM Proxy: one shared LLM gateway, a virtual key per client, separate budgets and bills
- ToolsBraintrust for FDEs: turning production failures into evals that block merges
- ToolsLangfuse: answering the three hardest questions after an LLM deployment
- Books & coursesEscaping the Build Trap: the book that teaches FDEs to measure their work by client outcomes, not feature counts
- Books & coursesAccelerate: four metrics FDEs can use to measure software delivery speed and stability
- Books & coursesHamel Husain and Shreya Shankar's AI Evals course: the most valuable lesson is reading the output
- AnalysisLangSmith, Arize, Helicone or PostHog: let the client's constraints choose the tool
Leadership
Run the project and feed lessons back to the product team.
Skills: Scope management · Working across teams
- GuidesHow to turn a finished deployment into a playbook, runbooks and a template repo for the next customer
- GuidesLeading engineering at a customer site when nobody reports to you
- Books & coursesSoftware Engineering at Google: the book that teaches FDEs to think in years, not lines of code
- Books & coursesThe Trusted Advisor: the book from 2000 that teaches FDEs how to win technical trust
- AnalysisHow many clients can one FDE carry? The answer depends on phase and tooling, not on a single number
- AnalysisDoes needing FDEs mean the product isn't finished? Follow the custom code
- AnalysisThe first FDE team: when to hire, what to measure and when to add an FDE Lead
Career
Build a portfolio and get through FDE interviews.
Skills: Portfolio · Interviews
- GuidesBuilding an FDE portfolio repo: a simple agent, evals built from real failures, a README and a 3-minute video
- GuidesTelling project stories in FDE interviews: how to answer "What trade-offs did you make?"
- Books & coursesRead Marty Cagan's Inspired and stop running a feature factory for your customers
- Books & coursesAlex Xu's System Design Interview: keep the four-step framework, add the AI layer yourself
- Books & coursesEducative's FDE course aims squarely at the interview loop, but only for engineers who have already built AI systems
- AnalysisWhen the word "resolved" decides an agent's revenue
- AnalysisThe FDE CV: drop the framework list, state deployment results with a baseline
- AnalysisWhen AI writes code at scale, FDEs are paid to own the outcome
- AnalysisFrom FDE to founder: which habits carry over and which must be relearned
- AnalysisAn FDE costs about $400,000 a year, so the model only scales if each deployment makes the next one cheaper
- AnalysisThree job postings, one ladder: FDE pay from one year of experience to eight