Self-assessment · 10 capability domains

Forward Deployed AI Engineer Skills and Portfolio Checklist

A credible FDE portfolio proves more than tool familiarity or an AI demo. It shows that you can discover a valuable problem, design and build the system, evaluate probabilistic behaviour, deploy with controls, drive adoption and measure impact. You need one strong engineering anchor and enough cross-functional fluency to defend the evidence behind a production decision.

How to score yourself

Use a four-point evidence scale. Do not average scores blindly: strong AI theory with no production software, security or stakeholder delivery still represents critical readiness gaps.

ScoreMeaningEvidence standard
0Not yet exposedCannot explain or demonstrate it
1UnderstandCan explain concepts and identify examples
2Apply with supportHas completed a guided lab or bounded task
3Demonstrate independentlyHas produced inspectable evidence and can defend trade-offs

The ten-domain checklist

1. Discovery & qualification

  • Map stakeholders, users, decisions and exceptions.
  • Separate a root problem from a feature request.
  • Establish a KPI baseline and measurement source.
  • Recommend “do not use AI” when appropriate.
Evidence: discovery brief, workflow, baseline, opportunity scorecard and go/no-go decision.

2. Business-to-technical translation

  • Write a testable problem and target outcome.
  • Define approvals, fallbacks and exception routes.
  • Create acceptance criteria and an evaluation plan before build.
Evidence: target workflow, backlog, decision records, evaluation and delivery plan.

3. Production software engineering

  • Build and review Python services and a modern interface or integration.
  • Design APIs, SQL data and failure handling.
  • Use tests, authentication, authorization and secure defaults.
Evidence: repository, tests, API contract, data model, access-control proof and documentation.

4. Data & knowledge engineering

  • Inventory sources, owners, classifications and quality.
  • Design governed ingestion, retrieval, lineage and freshness.
  • Evaluate retrieval independently from answer quality.
Evidence: source inventory, data flow, lineage, governed retrieval and retrieval evaluation.

5. LLM, RAG & agentic AI

  • Select models using quality, latency, cost and privacy.
  • Validate structured outputs and evaluate RAG.
  • Bound tools, state, permissions, approvals and escalation.
Evidence: behaviour contract, eval dataset, traces, tool schemas and failure analysis.

6. Enterprise integration

  • Connect APIs, events, SaaS and legacy systems safely.
  • Handle identity, delegated authorization, retries and idempotency.
  • Design for degraded dependencies and auditability.
Evidence: integration map, contracts, identity flow, failure tests and audit trail.

7. Evaluation & quality

  • Create representative, adversarial and regression cases.
  • Define metrics, segments, thresholds and release gates.
  • Calibrate human and model-assisted evaluation.
Evidence: dataset card, scorecard, taxonomy, calibration and release decision.

8. Security & governance

  • Threat-model prompt injection, leakage and excessive agency.
  • Apply least privilege, human oversight and audit controls.
  • Document residual risk and named accountability.
Evidence: threat model, access tests, risk register, approvals and incident controls.

9. Cloud & AI operations

  • Separate environments and gate release with CI/CD.
  • Instrument application, model and tool traces.
  • Rehearse rollback, fallback and incident response.
Evidence: pipeline, dashboards, alerts, rollback proof, runbook and incident review.

10. Adoption, value & communication

  • Run UAT around real roles and tasks.
  • Distinguish availability, adoption and value.
  • Present trade-offs to engineers, operators and executives.
Evidence: UAT, enablement, adoption dashboard, KPI report and executive readout.

Minimum viable FDE portfolio

One complete case study is stronger than six disconnected demos. Include the problem and baseline, current and target workflow, architecture, code evidence, governed data, evaluation dataset, threat model, CI/CD, rollback, observability, UAT, adoption, KPI evidence, individual contribution, failures and a reproducible verification guide.

Evidence quality test

Authentic: Did you materially contribute? · Inspectable: Can a reviewer verify more than a screenshot? · Contextual: Does it explain the decision? · Measured: Is there a baseline or threshold? · Safe: Is confidential information protected? · Defensible: Can you explain alternatives and failures?

Redaction and confidentiality

Never publish credentials, production data, personal information, proprietary code, customer names without permission, internal URLs or sensitive security architecture. Use synthetic data, redacted artifacts, private review or a reproducible simulation. A portfolio is evidence of judgment; exposing confidential information demonstrates poor judgment.

Interpret your result

PatternRecommended next step
Mostly 0–1 in engineering foundationsBuild programming, SQL, API, Git and cloud foundations
Strong anchor; weak AI/evaluationAdd focused applied-AI and evaluation work
Strong AI; weak software/operationsDeploy a tested service with CI/CD and observability
Strong technology; weak discovery/adoptionLead a stakeholder-facing deployment simulation or pilot
Mostly 2–3 with one clear anchorConsider advanced FDE preparation

This is developmental guidance, not an employment or admissions guarantee.

Next step

Turn your scores into a bridge plan.

Bring your strongest evidence and biggest gaps to a short advisor conversation.

Talk to an advisor →
Written by: Manikanta Kona · Reviewed by: KDigital Academic Quality Team · Last reviewed: 5 August 2026 · Evidence methodology · Sources: NIST AI RMF, Google Cloud production guidance.
At a glance

FDE Skills and Portfolio Checklist for Enterprise AI Roles

A ten-domain self-assessment for the engineering, delivery and evidence expected in a credible FDE portfolio.

What skills does an FDE need?

An FDE needs one deep engineering anchor plus working capability in discovery, software, data, applied AI, integration, evaluation, cloud operations, security, governance, adoption and communication.

Is a chatbot project enough for an FDE portfolio?

No. The portfolio must also prove governed data access, evaluation, security, integration, production operations, user workflow and measurable outcome.

How many projects should an FDE portfolio have?

There is no universal number. One complete, inspectable end-to-end deployment case is usually more credible than many shallow demos.