b01d5aa feat(builds): demo projects
Builds. The capstones from each track, written as demo projects. Finish the matching free lesson and its tile moves from Draft to Deployed (demo).
These are illustrative projects written by FoundyStack for this concept. They are not student work and have no real users. No star counts, no download numbers.
demo project · Agent Builder track Refund agent with limits Looks up orders, proposes refunds, and stops for a person above the limit.
Draft finish the lesson to ship it Architecture
Customer → message → Agent loop Agent loop → tool call → Validator Validator → valid args → Policy Policy → over limit → Approver Policy → within limit + key → Refund API Stack
TypeScript Tool schemas Approval queue Idempotency keys
policy.ts · ts Copy export function policy(args: RefundArgs): "run" | "approve" { return args.amount_cents > APPROVAL_LIMIT_CENTS ? "approve" : "run" ; } Tests
✓ rejects cut-off JSON ✓ holds 320.00 for approval ✓ never retries a write without a key Build this on the Agent Builder track demo project · App Engineer track Cited answer box Answers help-centre questions from retrieved paragraphs, with a link to each one.
Draft finish the lesson to ship it Architecture
Docs → markdown → Chunker Chunker → chunks → Index Question → query → Ranker Index → candidates → Ranker Ranker → top k above threshold → Model Stack
BM25 ranking Heading chunks Refusal threshold Recall at k
answer.ts · ts Copy if (!ranked.length || ranked[0 ].score < threshold) { return { answer: "I don't have enough material to answer that." , citations: [] }; } Tests
✓ refuses questions outside the docs ✓ recall at 3 >= 5 of 6 ✓ every answer has a citation Build this on the App Engineer track demo project · Production track Trace and cost dashboard One page showing traces, cost per run, latency per step and refusal rate.
Draft finish the lesson to ship it Architecture
App → spans → Tracer Tracer → redacted attributes → Store Store → queries → Dashboard Store → error, cost, refusal rates → Alerts Stack
Spans Token counts Latency budgets Alerts
span.ts · ts Copy span.set({ model: m, tokens_in, tokens_out, cost_usd: price(m, tokens_in, tokens_out) }); span.set({ user_text_len: text.length }); // shape, not content Tests
✓ every request has one trace id ✓ cost per run under budget ✓ no raw user text stored Build this on the Production track demo project · Launchpad track Inbox triage helper Sorts a week of emails into buckets and drafts replies a person approves.
Draft no free lesson yetArchitecture
Inbox export → rows → Script Script → one item → Model Model → bucket + draft → Validator Validator → approve or edit → Person Stack
Spreadsheet export Short script Structured output Approve button
item.json · json Copy { "bucket" : "finance" , "needs_reply" : true , "draft" : "Thanks, we will pay invoice 4471 on Friday." } Tests
✓ bucket is one of 5 values ✓ nothing sent without approval ✓ a week of misses logged Build this on the Launchpad track demo project · App Engineer track Voice intake summariser Turns recorded call transcripts into structured tickets with a confidence flag.
Draft no free lesson yetArchitecture
Call recording → speech to text → Transcript Transcript → text → Model Model → ticket JSON → Validator Validator → low confidence to a person → Queue Stack
Transcripts Ticket schema Low-confidence queue
ticket.ts · ts Copy type Ticket = { topic: "billing" | "delivery" | "account" | "other" ; summary: string; confidence: "high" | "low" };Tests
✓ topic is an enum ✓ low confidence goes to a person ✓ summary under 280 characters Build this on the App Engineer track demo project · AI Systems track Build or buy decision record A one-page record scoring a vendor and a prototype on the same thirty cases.
Draft no free lesson yetArchitecture
Real cases → 30 examples → Test set Test set → same inputs → Vendor Test set → same inputs → Prototype Scores → plus cost and risk → Decision Stack
Test set Bake-off sheet Cost estimate Rollback owner
decision.md · md Copy ## Decision Buy for now. Vendor scored 26 /30 , prototype 21 /30 . Revisit on the first of next quarter. Owner who can switch it off: support lead. Tests
✓ cost of a wrong answer written down ✓ both options scored on the same cases ✓ named owner for rollback Build this on the AI Systems track