FDE Capstone and 30/60/90 Plan for Enterprise AI

At a glance

Figure 1. The capstone service in one view

How to read this: every request passes through the Orchestrator, which verifies the caller before any retrieval or tool call.

You build a Python service that turns a vague request into requirements and cited recommendations.

Eight phases, P1 to P8, sit inside modules M1 to M6 and take about 4.5 hours.

It shows you can build and measure such a service. It does not guarantee a job.

Scenario

Composite scenario, not a real company. Lumen Industrial Supply, a fictional distributor, has sales engineers who answer requests such as "monitor vibration on 40 pumps in a humid plant, around 30k, must work with our OPC UA setup" by searching a catalog by hand. A first version should extract requirements, find matching products with the supporting passage, say what it cannot answer, and never invent a product. Numbers are illustrative.

Core ideas

What the Copilot does

Input	Output
A free-text request and a role (public or partner)	Structured requirements (category, quantity, budget, protocols, environment)
A catalog of 40 fictional products and 30 product documents	Recommendations citing the passage
Unanswerable or ambiguous questions	A refusal or a clarifying question

Three behaviours are measured: it cites evidence, it refuses when the documents do not support an answer, and a public user never sees partner-only content.

At the customer: "How will you know it works?" Name the evaluation: a hand-written question set, including questions that must be refused, and its results.

The eight phases

Figure 2. Eight phases hosted by six modules

flowchart LR accTitle: Eight phases hosted by six modules accDescr: Module M1 hosts phases P1 requirements and dataset and P2 API and deterministic search. M2 hosts P3 requirement extraction. M3 hosts P4 retrieval and evidence. M4 hosts P5 bounded tool workflow. M5 hosts P6 validation, access control and errors. M6 hosts P7 tests and evaluation and P8 UI, documentation and demo. subgraph M1["M1"] P1["P1 Requirements and dataset"] --> P2["P2 API and search"] end subgraph M2["M2"] P3["P3 Requirement extraction"] end subgraph M3["M3"] P4["P4 Retrieval and evidence"] end subgraph M4["M4"] P5["P5 Bounded tool workflow"] end subgraph M5["M5"] P6["P6 Validation and access control"] end subgraph M6["M6"] P7["P7 Tests and evaluation"] --> P8["P8 UI and demo"] end P2 --> P3 --> P4 --> P5 --> P6 --> P7

How to read this: each phase needs the one before, so the fast track still builds P1 to P3.

Phase	Work	Module	Hours
P1	Requirements and dataset	M1	0.4
P2	API and deterministic search	M1	0.7
P3	LLM requirement extraction	M2	0.5
P4	Retrieval and evidence	M3	0.7
P5	Bounded tool workflow	M4	0.7
P6	Validation, access control and errors	M5	0.4
P7	Tests and evaluation	M6	0.5
P8	UI, documentation and demo	M6	0.6
	Total		4.5

What the rubric will reward

The capstone rubric scores 100 points across seven criteria:

Correctness and tests: pass offline, cover failure paths.
Retrieval and evidence: real citations; unanswerable questions refused.
Workflow safety: step and time limits, read-only tools, an approval gate.
Security and access control: no partner-only content for a public user.
Evaluation honesty: you ran the 36-question golden set (6 unanswerable, 4 access-control, 4 ambiguous).
Integration design note: one to two pages.
Documentation and demo: the README runs from a clean checkout.

The 30/60/90 template

30/60/90 PLAN
Hours per week I can protect: ____
Overrun factor over the course plan (start with 1.5): ____
Path: full / fast        Review date: ____

DAYS 1-30   Goal: ____
  Lessons: ____    Capstone phases: ____
  Hours needed (plan hours x factor): ____   Hours available: ____
DAYS 31-60  Goal: ____
  Lessons: ____    Capstone phases: ____
  Hours needed: ____   Hours available: ____
DAYS 61-90  Goal: ____
  Lessons: ____    Portfolio and outreach: ____
  Hours needed: ____   Hours available: ____
Risks and what I will cut first: ____

Worked example

Assumptions (illustrative): 4 protected hours per week, an overrun factor of 1.5, the full path.

Block	Content	Plan hours	Hours needed (x 1.5)	Hours available	Spare
Days 1-30	M0 to M3, P1 to P4	8.8	13.2	17.1	3.9
Days 31-60	M4 to M6, P5 to P8	6.2	9.3	17.1	7.8
Days 61-90	M7 to M10	4.5	6.75	17.1	10.4

Check: block one is 0.5 + 2 + 2 + 2 (M0 to M3) plus 0.4 + 0.7 + 0.5 + 0.7 (P1 to P4) = 8.8. Available hours are 30 / 7 x 4 = 17.1. The blocks total 19.5 of 20 hours.

Practice task

Copy the template into a file. Write your weekly hours and overrun factor, fill the three blocks with lessons and phases, and compare needed with available hours. Where the spare is negative, cut work, and note what you cut first.

Exercise

Explain it back. In three sentences, explain the capstone to someone new. Cover what goes in and comes out, how you know it works, and what it refuses to do.

Done when all three blocks show needed against available hours and your three sentences are written.

Hint 1: too optimistic

Raise the overrun factor to 2 and recompute.

Hint 2: only a feature list

Lead with the problem: "sales engineers lose hours matching requests to datasheets."

Common mistakes

Planning with no overrun factor.
Putting optional lessons in the first block.
Treating the capstone as a demo; the evaluation is the proof.
Skipping P1 and P2 on the fast track.

Knowledge check

1. Which three behaviours matter most?

It cites evidence, refuses when the documents do not support an answer, and hides partner-only content from public users. Each is measured.

2. Why eight phases hosted inside modules?

Each phase uses the concept its module teaches. Each builds on the one before.

3. Why use an overrun factor when you plan?

Course hours assume focused work; first attempts run long. A stated factor shows it.

4. How many hours do P5 to P8 take, and which module hosts each?

P5 is 0.7 in M4, P6 is 0.4 in M5, P7 is 0.5 and P8 is 0.6 in M6, totalling 2.2 hours.

5. Does finishing guarantee production-readiness?

No. It uses fictional data, a mock provider and a small question set. Production needs real data and a security review.

You can now

Describe the capstone's inputs, outputs and refusals.
Name the phase each of M1 to M6 hosts.
Fill in a 30/60/90 plan.

FDE Capstone and 30/60/90 Plan for Enterprise AI

At a glance

Scenario

What you will be able to do

Prerequisites

Core ideas

What the Copilot does

The eight phases

What the rubric will reward

The 30/60/90 template

Worked example

Practice task

Exercise

Common mistakes

Knowledge check

You can now

Further reading

Next

Enroll to keep reading

Ask about this article

Enjoyed this? Get more like it
every Monday.

Comments

At a glance

Scenario

What you will be able to do

Prerequisites

Core ideas

What the Copilot does

The eight phases

What the rubric will reward

The 30/60/90 template

Worked example

Practice task

Exercise

Common mistakes

Knowledge check

You can now

Further reading

Next

Enroll to keep reading

Ask about this article

Enjoyed this? Get more like it every Monday.

Comments

Enjoyed this? Get more like it
every Monday.