Veredas

Aspiring AI engineers in Latin America face a paradox: they know they want to work with artificial intelligence, but they don't know exactly which skills they're missing, in what order to learn them, or what salaries and roles actually exist in their region. Available information online is mostly centered on the US market and in English, creating an additional gap for those starting from Argentina, Colombia, Mexico, or any other country in the region.

VEREDAS takes the user's current skill set, runs a gap analysis using topological sort with DFS to guarantee the learning order always respects skill dependencies, and generates a personalized roadmap. Claude AI produces honest, actionable guidance in the user's language. The interactive skill map uses physics simulation with vis.js. Latin American market data comes from curated sources including Glassdoor, remoteok.com, and levels.fyi. Final project for Stanford Code in Place 2026.

• 01

  Implementing a recursive DFS topological sort that guarantees the learning order always respects skill dependencies without cycles or inconsistencies

• 02

  Integrating real Latin American market data from a curated CSV using Python's csv module, cross-referencing it with real-time data from the GitHub REST API and ExchangeRate API

• 03

  Building a complete internationalization system in vanilla JavaScript without frameworks, with seamless ES/EN toggle across all components including Claude's output

• 04

  Rendering an interactive dependency graph with physics simulation in vis.js that remains readable and usable on mobile without losing the ability to explore nodes and dependencies

• 05

  Generating a completion certificate in PDF with html2pdf.js that activates automatically when the skill gap reaches zero

• 01

  Topological sort isn't just an academic algorithm: applied to a real skill graph, it produces a learning order no human could manually calculate for 20+ skills with cross-dependencies

• 02

  Building without frameworks forces you to understand exactly what's happening in every DOM interaction, producing more predictable and debuggable code than previous versions of the same project with React

• 03

  Bilingual prompting with Claude requires explicit language instructions in the system prompt, not just the user message, for consistent output regardless of input language

• 04

  Latin American market data is scarce and scattered: curating it manually in a CSV was a product integrity decision, not a technical limitation

• 05

  Stanford Code in Place teaches Python from scratch with rigor: the project demonstrates recursion, classes, File I/O, REST APIs, and CSV parsing in a real functional product, not isolated exercises