# Capstone Spec Sheet
One-page spec for your VetsLearn capstone.

_The single-page spec sheet for an Intermediate Tier capstone project. Forces a real user, measurable success criteria, and a demo plan._

Used in: Lesson intermediate/10 — https://vetslearn.com/lessons/intermediate/10
Source: https://vetslearn.com/resources/templates/engineering#t-capstone-spec-sheet

---

# Capstone Spec Sheet: [PROJECT NAME]

**Owner:** [OWNER]
**Date:** [DATE]

---

## 1. Project name

*Fill this in: one short name. No tagline. No "AI-powered."*

## 2. Problem

*Fill this in: the problem in two sentences. Who has it. What it costs them today.*

## 3. User

*Fill this in: one named user persona. A real person if you can name one. "Sarah, a benefits counselor at the VA who reads 40 claims a day" beats "end users."*

## 4. Success criteria

*Fill this in: 3 measurable criteria. Each one has a number.*

1. [Criterion 1 — e.g. cuts time-per-claim from 12 min to under 5 min on a test set of 50]
2. [Criterion 2 — e.g. 90% pass rate on the eval set]
3. [Criterion 3 — e.g. under $0.05 per claim at projected volume]

## 5. System diagram

*Fill this in: a sketch. Boxes for user, app, model, retrieval, tools, logging. Arrows for data flow. Hand-drawn is fine.*

```
[user] -> [app] -> [orchestrator] -> [model]
                       |
                       +-> [retrieval]
                       +-> [tools]
                       |
                       v
                   [logging]
```

## 6. Eval plan

*Fill this in: where the eval set comes from, how big, how it gets scored, who reviews.*

- Source: [text]
- Size at launch: [N] cases
- Tags: happy / edge / adversarial
- Scoring: [exact match / LLM judge + human review]
- Re-run cadence: [every PR / nightly / weekly]

## 7. Cost estimate

*Fill this in: math, not vibes. Tokens per call x calls per day x price.*

| Item | Per call | Daily volume | Daily cost | Monthly cost |
|---|---|---|---|---|
| Model | [$] | [N] | [$] | [$] |
| Embeddings | [$] | [N] | [$] | [$] |
| Vector DB | [$] (flat) | — | — | [$] |
| Hosting | [$] (flat) | — | — | [$] |
| **Total** | | | | **[$]** |

## 8. Timeline

*Fill this in: weekly milestones. Be honest about scope.*

- Week 1: [milestone]
- Week 2: [milestone]
- Week 3: [milestone]
- Week 4: [milestone — demo]

## 9. Risks

*Fill this in: top 3 things that could sink this. Name the mitigation for each.*

1. [Risk] — mitigation: [text]
2. [Risk] — mitigation: [text]
3. [Risk] — mitigation: [text]

## 10. Demo plan

*Fill this in: what you show, in what order, for how long.*

- Minute 0-1: the problem, the user, why it matters
- Minute 1-4: live walkthrough — happy path, edge case, one failure mode handled gracefully
- Minute 4-5: eval pass rate, cost, next step

---

## How to use

One page is the point. If a section grows past a few lines, you are designing, not specifying. Walk through this with one other person before you start building. Pin it in the repo README. When scope creep hits, this is the document that says "not in v1."

## Check before you use it

- Each success criterion has a number you can measure with what you have.
- You talked to at least one person who matches the user section.
- The cost math comes from token counts on a real sample run, not from list prices alone.
- It still fits one page.