Personal Projects

Building ideas with impact - from AI memory agents to learning platforms and developer ecosystems.

These projects blend machine learning, thoughtful engineering, and human outcomes. I like to take an idea from a personal need, design the stack, ship a working product, and then refine it until it feels genuinely useful.

Projects

Four builds that mix AI, product craft, and purposeful storytelling. No links of github because they are private, but any concerns, feel free to reach out.

Talk With Loved Ones

AI memory-preservation agent built with ElevenLabs voice cloning + RAG + STT/TTS

A personal build to preserve the voice and stories of my grandmother, Macita. Family members can speak with an AI agent that answers in her cloned voice, retrieving memories, teachings, and personality details without trying to replace her.

  • ElevenLabs-based voice cloning using curated audio of my grandmother.
  • RAG pipeline over vectorized memories, stories, and personal documents.
  • Speech-to-text and text-to-speech loop for natural conversation.
  • Detects who is speaking and loads tailored memories for that person.
  • Backend built with Python and FastAPI with structured memory stores.
  • Modular, privacy-focused design to keep sensitive family data secure.
Voice Cloning RAG STT/TTS Personal AI Agents Memory Preservation FastAPI Python

This project was built out of love - an engineering way to keep family stories alive.

Looqs

Fashion recommendation engine powered by LLM agents

An AI fashion assistant that analyzes profiles, colorimetry, and preferences to generate curated outfits. Looqs orchestrates LLM agents to pair shoppable looks with real inventory scraped from retailers.

  • Multi-agent workflow using LangGraph and OpenAI Assistants.
  • Outfit generation tuned to mood, occasion, and color palette.
  • Product enrichment agent scraping real retail inventory (Palacio de Hierro).
  • Supabase integration for identity, preferences, and session continuity.
  • Scalable backend built with Python and Flask.
LLMs LangGraph Fashion AI Web Scraping Supabase Flask Personalization
GitHub

Gradooz

Adaptive learning platform for UNAM applicants

A Duolingo-style prep app built to democratize UNAM admission readiness. Around 200,000 students take the exam each year and fewer than 20,000 get in; Gradooz delivers structured practice, adaptive difficulty, and feedback to close that gap.

  • Interactive quiz engine that adapts to user performance.
  • Large question bank organized by subject and difficulty.
  • Real-time analytics to track progress and focus areas.
  • Motivational gamification with streaks, achievements, and goals.
  • Designed mobile-first so learners can practice anywhere.
EdTech Adaptive Learning Python Mobile-first UI Exam Preparation

2Sunrise

Developer training and placement for global startups

Nearshoring engine connecting global companies with high-potential Mexican developers. 2Sunrise trains candidates in technical skills and communication so they are ready for international collaboration.

  • Training pipeline combining ML/AI skills with communication coaching.
  • Matching engine that pairs developers with US and global startups.
  • Multi-stage evaluation covering technical depth and culture fit.
  • Career acceleration approach to unlock senior-level opportunities.
Nearshoring Developer Training Talent Matching Communication Coaching Startups