Forward Deployed Engineer (FDE): Applied AI Delivery
Turn an AI-assisted workflow into a testable, permission-aware service with explicit human approval, dependable integrations and an operational handoff.
What you'll be able to do
- Translate an ambiguous assistant request into measurable acceptance criteria.
- Validate API contracts and make retry behavior explicit.
- Separate configuration, secrets and deployment trust boundaries.
- Filter document evidence by the requesting user's permissions.
- Persist approved workflow states and handle stale approvals.
- Adapt legacy responses without trusting unverified identity claims.
- Evaluate service reliability and assemble an actionable handoff.
Before you start
- Write small Python functions and read JSON payloads.
- Understand HTTP requests, responses and basic authentication concepts.
- Run Python 3.12 or newer locally; all core labs use the standard library and synthetic data.
The curriculum
- Turn a Request into a Delivery Contract — Free preview
Define permitted actions, exclusions and evidence for a successful release.
- Build an API Contract That Can Fail Clearly — Free preview
Make validation, error responses and retry expectations explicit before connecting a model or an external ticket system.
- Package the Runtime and Its Trust Boundaries — Sign-in access
Keep configuration reproducible, secrets out of artifacts and cloud permissions narrower than the service's code surface.
- Retrieve Only Evidence the Requester May Read — Sign-in access
Filter evidence before selection and cite only authorized material.
- Make Approval a Durable State, Not a Button — Free preview
Bind a review decision to an exact draft revision and reject stale actions after edits or concurrent changes.
- Bridge Identity and Legacy Systems Without Borrowing Trust — Sign-in access
Translate older system formats through a narrow adapter while keeping identity, access and status semantics explicit.
- Measure the Service and Rehearse Recovery — Sign-in access
Define meaningful success metrics, inspect tail latency and create a failure drill that produces a usable operator handoff.
- Assemble Relay Room and Defend the Release — Sign-in access
Integrate the course boundaries into a local portfolio project and present evidence without overstating production readiness.
Relay Room
Build a local facilities service-request assistant for fictional shared offices. It helps staff draft requests about furniture, cleaning and presentation equipment using permitted policy notes. A human reviews the precise draft before a simulated ticket is created. Emergency response, autonomous purchasing and production identity integration are outside the project scope.
Deliverables
- A runnable standard-library Python reference workflow and synthetic fixtures.
- An acceptance matrix and API contract tests.
- A permission-filtered document lookup and trusted identity adapter.
- A persistent approval ledger with a fake idempotent ticket adapter.
- A small reliability report, failure drill and operator handoff.
Review your work
- Unauthorized documents never appear in a response or model-facing context.
- A changed draft invalidates its previous approval.
- A retry of an approved operation produces one simulated ticket.
- Malformed upstream data is rejected without exposing internal details.
- The demo runs locally without real employee data, paid APIs or cloud accounts.