Abril Care Agent

Parents miss 13 workdays/year to childcare gaps. Abril searches 24K+ providers, spots coverage gaps from your calendar, and texts you backup care before you ever scramble.

Comment

Inspiration

Parents miss an average of 13 workdays per year due to childcare breakdowns. The childcare crisis costs the U.S. economy $122 billion annually. And 71% of that invisible coordination labor — the mental load of planning, scheduling, and scrambling — falls on mothers.

I built Abril because current tools (Care.com, Winnie, Kinside) are fundamentally reactive. They offer directories for parents to scroll through after they are already in crisis. Abril is different: it's an operating system for the care village. It spots coverage gaps weeks in advance and texts you with backup care options before you ever need to scramble.

Named after the Spanish word for April — the month of new beginnings — Abril is designed to be the proactive care coordination partner every parent deserves but never had.

What it does

Abril is an AI care coordination agent built on DigitalOcean Gradient AI that helps families in three ways:

1. Intelligent Provider Search (RAG) Search 24,394 verified childcare providers across the DC-Maryland-Virginia metro and New York City metro areas. Each result is scored using Haversine distance, normalized quality ratings across 4 state systems (VA VQB5, DC Capital Quality, MD EXCELS, NY OCFS), age-appropriate matching, and capacity. This isn't keyword search — it's semantic RAG retrieval through DigitalOcean's Knowledge Base.

2. Proactive Gap Detection Abril cross-references school district calendars (MCPS, FCPS, DCPS, NYC DOE) against your work schedule to identify coverage gaps — school closures, early dismissals, professional development days — before they become emergencies. Each gap is severity-scored so you know what to tackle first.

3. SMS Alerts via Twilio Six weeks before spring break, Abril texts you with specific backup care suggestions. That's the difference between scrambling on Sunday night and having a plan on Monday morning.

4. Multi-Child & Budget Awareness Families with multiple kids can search for providers that serve all their children's age groups at once, with estimated combined monthly costs and budget filtering.

5. Emotionally Intelligent & Equity-Minded Persona Abril is warm but not patronizing, equity-minded (frames recommendations for both parents), culturally fluent across 4 states/regions, and remembers your child's name, age, and preferences across the conversation.

How we built it

Abril is built entirely on DigitalOcean — both the AI agent and the production app. The live web application runs on DigitalOcean App Platform, auto-deploying from GitHub on every push, while the AI backend leverages the full Gradient AI platform:

  • Agent Development Kit (ADK): 3 @entrypoint functions — main() for chat, connect_calendar() for Google Calendar OAuth + gap detection, and send_alert() for Twilio SMS.
  • Knowledge Base (RAG): 24,394 provider records indexed with OpenSearch + GTE Large EN v1.5 embeddings. Hybrid search (alpha=0.5) balances lexical matching with semantic understanding.
  • Serverless Inference: Claude 3.5 Sonnet via DigitalOcean's inference endpoint. Conversation memory maintains last 10 turns for multi-turn context.
  • Guardrails: Code-level PII protection (blocks SSN, credit card, bank account patterns), email redaction, and auto-appended medical disclaimers.
  • Tracing: @trace_retriever and @trace_llm decorators for full observability.
  • Evaluations: Test cases in config.yaml validate provider search accuracy, gap detection, and response helpfulness.
  • App Platform: The production web app is deployed on DigitalOcean App Platform with automatic GitHub deploys, giving Abril a fully DigitalOcean-native stack from hosting to inference.

Additional integrations: Twilio for SMS alerts, Google Calendar API (OAuth2) for real calendar-based gap detection, and 5 public datasets merged with a custom quality crosswalk algorithm.

Challenges we ran into

  • Quality crosswalk: Four different states rate childcare differently — Virginia uses VQB5, DC uses Capital Quality, Maryland uses EXCELS, and New York uses CMS Stars. Normalizing them to a single 0-100 scale required domain research and careful calibration.
  • Knowledge Base scale: Indexing 24,394 JSONL records with rich metadata required careful chunking and metadata design to ensure hybrid search returned relevant results.
  • Multi-child complexity: Building a scoring system that requires ALL children's age groups to be served by a single provider, while estimating combined monthly costs, required rethinking the search pipeline.

Accomplishments that we're proud of

  • 24,394 verified providers across 2 major metro areas, with quality ratings normalized across 4 different state systems
  • An agent persona (Abril) that feels like a trusted friend, not a search engine
  • Proactive SMS alerts that notify families about gaps weeks before they happen

What we learned

That the childcare coordination problem is fundamentally a data integration and prediction problem, not a search problem. Parents don't need another provider directory — they need someone who connects their calendar, their school district's calendar, and provider availability into a single view and tells them what to do about it.

DigitalOcean Gradient AI made it possible to build this as a hackathon project because the ADK, Knowledge Base, and inference are all integrated. One platform handled agent hosting, retrieval, inference, guardrails, and observability.

What's next for Abril

  • National expansion: Starting with the top 10 metro areas
  • Employer partnerships: B2B model where employers offer Abril as a benefit
  • Real-time calendar sync: Always-on gap monitoring with push notifications
  • Care village coordination: Invite grandparents, nannies, and co-parents into the system

Built With

Share this project:

Updates