Learn · the main route
The FDE path.
35 lessons in 6 stages, in the order a real client project meets the work: watch the work and find the data first, then one model call you can defend, then retrieval, agents, proof, and shipping. About 60 h of reading and hands-on work.
Every stage runs in one order: Learn its lessons, Practise in a lab placed right after the lesson it practises, Apply it in a Halden week, then a Checkpoint drill. A stage lists only the parts it has, and ends with the last of them: Stages 1 and 5 with a checkpoint drill, Stages 2 and 3 with its lessons, Stage 4 with a lab, Stage 6 with Halden weeks. Work each stage top to bottom. Up next marks your next lesson; a stage’s Halden weeks and checkpoint come once its lessons are done.
- Lab
- A client's messy data with known problems: write Python in the browser to find them and get a score, graded here.
- Client data (estate)
- A fictional client's files, with documented problems planted in them. Labs, Halden weeks and role projects all run on one.
- Halden week
- One week with your client, Halden Freight, the simulated company of the residency: something changes, you do the work in your own repository, and it's graded on your machine.
- Checkpoint drill
- A 45-minute mock design round, with a rubric to score yourself. A checkpoint drill checks one stage.
- Specialisation
- Extra lessons, a role project and drills for one kind of FDE role, taken after the path.
- Role project
- A complete engagement with a second client for one kind of role, from brief to sign-off.
Aiming at one kind of role? Add a specialisation. Looking for one topic? Everything is in the lesson library.
Stage 1
The first two weeks on site
Watch the work, find the real requirement and the data, and decide what kind of system this is before you build anything.
This stage: LearnPractiseApplyCheckpointFree
0 of 6 understood · 6 lessons · 1 lab · 4 Halden weeks · 1 checkpoint drill · 10.8 h
- Shadowing the work as it's really doneUp next
- Finding the real requirement
- Finding where the data really lives
- Lab Load the desk's spreadsheet
- Feasibility and data access, fast
- The operational model: objects, links and actions
- Rules, SQL, models, solvers or an LLM?
- Halden week 1 Access stalls; only an export
- Checkpoint drill First week: discovery with a sceptical operations manager Pro
- Later Halden weeks that use this stage
- Role projects
- Interview drills
Stage 2
One trustworthy model call
Make a single model call you can defend: what it reads, what shape it answers in, and what it can cite.
This stage: Learn Pro
0 of 3 understood · 3 lessons · 4.5 h
Stage 3
Answers from their data
Get the customer's documents and tables into answers, with the permissions they already have.
This stage: Learn Pro
0 of 4 understood · 4 lessons · 7.4 h
Stage 4
Agents that act safely
Let the system change things in their systems, with gates, retries and a screen the operator trusts.
This stage: LearnPractise Pro
0 of 7 understood · 7 lessons · 1 lab · 11.5 h
- The agent loop, written by hand Pro
- Agent or workflow? Choosing the autonomy level Pro
- Designing tools a model can use well Pro
- Approval gates and bounded autonomy Pro
- Retries, fallbacks and failing gracefully Pro
- Lab Find the intake agent's failed writes Pro
- Agents inside CRM, ERP and ITSM Pro
- The screen the operator uses all day Pro
- Later Halden weeks that use this stage
Stage 5
Prove it and pass review
Measure before you claim, gate every change, and survive the security review.
This stage: LearnPractiseApplyCheckpoint Pro
0 of 7 understood · 7 lessons · 1 lab · 2 Halden weeks · 1 checkpoint drill · 12 h
- From requirements to eval criteria Pro
- Golden datasets and the metrics that matter Pro
- Lab Clean a golden set before you trust it Pro
- LLM judges and rubrics you can trust Pro
- Regression tests and CI gates Pro
- Acceptance criteria and sign-off Pro
- Defending against prompt injection Pro
- Threat modelling an AI system Pro
- Checkpoint drill A support agent got worse after a model upgrade: find out why by Friday Pro
- Role projects
Stage 6
Ship, run and hand over
Roll out in stages, see what it does, keep it within budget, run the engagement, reach the next site, and leave.
This stage: LearnPractiseApply Pro
0 of 8 understood · 8 lessons · 1 lab · 4 Halden weeks · 13.8 h
- Canary releases and staged rollout Pro
- Tracing agents with OpenTelemetry Pro
- Lab Check a trace export before you trust the dashboard Pro
- SLOs and latency budgets Pro
- Cost per resolved case Pro
- Scoping a shippable slice Pro
- Rhythm, writing and people on site Pro
- From one site to many Pro
- Handover, run state and the next use case Pro
- Role projects
- Exercise 2. Accessibility acceptance criteria and an automated check · Kelverne client project
- Capstone: launch readiness review with eval results and the ask · Doorvale client project
- Exercise 4. Incident: the portal shows wrong payment dates for six hours · Kelverne client project
- Capstone: authorisation evidence pack and go-live decision memo · Kelverne client project