AI · Health insurance
BuscoPrepaga
Helping people in Argentina compare health insurance through a conversation grounded in plan data.
Describe your needs
Describe your age, location and budget in your own words.
Real product walkthrough. Typing and waiting times shortened for viewing.
The Vision
Help each person find the health insurance plan that fits all their needs and requirements through a single conversation—without contacting providers one by one to request prices and personalized quotes.
My contribution
My work spans product design, full-stack development, the AI assistant, and infrastructure.
- Status
- Final stage of development
- Year
- 2026
- Role
- Co-founder
The problem
In Argentina, private health insurance plans are known as prepagas. Comparing them means contacting every provider one by one, requesting a personalized quote, and then comparing the results by hand.
Each conversation also becomes a sales pitch. That makes an already opaque process exhausting for someone who simply wants to understand their options.
The product
BuscoPrepaga brings plan pricing and coverage into a searchable comparison. People can filter their options or talk to Sofía, the AI assistant.
Sofía asks about age, location, budget, and coverage needs, then uses that profile to recommend matching plans. A price guard checks figures against the database before the response reaches the user.
An importer built against Argentina’s official health-insurance registry supports the underlying dataset. Plan pricing is checked against the application database.
Decisions that matter
Official data, not scraped quotes
Prices come from the public tariff filings that insurers submit to Argentina’s health regulator (SSSalud). They’re imported every month and mapped to each plan by one pure resolver that every price goes through. Every number on the site can be traced back to an official filing.
The model talks, the database decides prices
Sofía, the AI advisor, calls the Anthropic API directly with tool use. The tools quote plans, look up coverage, calculate amounts and capture leads. The model never makes up a figure: it gets prices from the tools, and a deterministic guard checks every amount in its reply against the database before the user sees it.
One pricing engine for everything
People can explore plans through a conversation with Sofía or browse the comparison pages, choosing whichever feels more comfortable. Both options share the same pricing engine, adapted to each experience, so the same plan and personal profile always produce the same price—regardless of how someone searches.
How it’s built
Data pipeline
A cron job imports the latest SSSalud filings, and six detectors run after each import: new plans with no mapping, regions with no mapping, prices that break the expected age curve, and more. Findings go to an admin inbox and an email, so data problems are caught before a user runs into them.
How a quote is built
Sofía doesn’t come up with prices herself. The model reads the user’s details (location, ages, payment type, budget) and passes them to a quoting tool. If a detail is missing, it’s left empty instead of guessed, and Sofía asks for it. The tool finds each insurer’s pricing zone, matches the official SSSalud rates to the catalog, filters by budget and calculates the totals. A final check blocks any reply with an amount that didn’t come from a tool result.
Testing an AI advisor
- Pure unit tests cover the parsing and the pricing engine, plus a frozen fixture that locks the plan resolver’s output.
- Sofía is also tested against a bank of scripted conversations run on the real model, each one evaluated automatically. In addition, manual tests cover rare cases devised by the founders to make the assistant truly reliable.
- Location parsing was measured on hundreds of hard cases, including towns with the same name in different provinces and surnames that match neighborhoods.
Under the hood
My scope
- Product design
- Frontend
- Backend
- AI architecture
- AI assistant testing
- Data engineering
- Infrastructure
Technology
- Next.js 16
- React 19
- TypeScript
- Supabase
- Anthropic
- AWS
- Claude Code & Codex
Explore the workflows
A conversation with Sofía, from profile to plan selectionFind a plan with Sofía
A conversation with the AI assistant, from a personal profile to a plan shortlist.
Share your profile with Sofía
Start a conversation with your age, location and budget.
Real product walkthrough. Typing and waiting times shortened for viewing.