Ale Cabrera
SHIPPED · DPARENTS · 2026 · 10 WEEKS
Building a Source-First AI Agent for Traveling Families
An AI search agent for worldschooling families that plans its own search, checks what it found, and only cites sources it can actually show.
- 75%of 163 hand-rated results judged accurate, 2 hallucinated
- 1,330questions traced end to end across 291 sessions and 4 countries
- 20real users gave feedback, two product changes shipped from it
163 results rated by hand
Accurate: 122
Partial or off-target: 36
Needing sources: 3
Hallucinated: 2
Catalog review, 5,668 entries live
Regulations human-verified: 100% (270)
Entities human-verified: 81% (626 of 777)
Resources AI pre-screened: 72% (2,857 of 3,954)
How the agent answers
- Plan
pick tools + queries - Search
catalog, web, YouTube - Evaluate
good enough? - Finalize
grounded answer + sources
Three risks, and what happened
| Risk | How it was handled | Result |
|---|---|---|
| Value. Generic web and YouTube search returns a wall of results with no judgment about a specific family. Without trust there is no product. | Trust as a data problem first: a structured taxonomy with review states, seeded by a hand-curated set of 300 entries. | Every source shown in an answer is a real search hit, never a model-invented URL. |
| Feasibility. Unclear whether a solo, AI-built agent could stay grounded. | A LangGraph agent that plans, searches, evaluates, then refines or finalizes. Refinement capped at 2 rounds. | 75% of 163 hand-rated results accurate, 2 hallucinated. |
| Compliance. Logging real questions with location context meant retention and privacy had to be designed in, not added later. | Request-log IPs stored as HMAC hashes, 90-day retention with a daily purge. | Image uploads only with consent. |
My role. Defined the problem and the trust principle the product is built on. Hand-curated the original dataset of 300 entries. Designed the taxonomy and the review states. Owned the guardrail spec.
AI collaboration. Claude Code and Cursor stood in for an engineering team: 220 merged PRs, 145 database migrations, 249 automated tests.
Product leader with a data scientist past · Zapier, Clearbit, HubSpot
Product doc: ale-cabrera.com/portfolio/dparents