😄 I'm Alonso Astroza Tagle. This is my resume.
- I'm the Deputy Director of Industry Partnerships @ Data Science Institute of Universidad del Desarrollo.
- Independent Contributor: GeoVictoria - Defontana - Discolab - Subconscious.ai.
- Lecturer of the Data Science Master’s program @ Universidad del Desarrollo.
- Electrical Engineer @ Universidad de Chile.
- [June 2025] 📺 Built tvtxt, an experimental AI pipeline that watches live TV streams and converts them into movie scripts in real-time. The system combines NVIDIA's Parakeet ASR for speech transcription with Qwen2-VL for visual scene analysis, using Outlines to enforce proper screenplay JSON formatting. Running on Modal's cloud infrastructure, it processes live broadcasts and outputs formatted screenplays with no storage or history, each moment exists only in the present, like live theater. A wild proof-of-concept that demonstrates real-time multimodal AI capabilities.
- [December 2024] 🚀 Founded IADevs, a software professional community focused on AI solutions with the motto "We don't talk about the future, we build it." Rather than discussing AI "possibilities" or the "future" of technology, we convert ideas into operational solutions. The community brings together developers who are actively implementing AI in production environments, sharing practical knowledge and real-world applications.
- [August 2024] 🍲 Carbonada is an AI-powered educational game that helps users, especially children, learn about their carbon footprint in a fun, interactive way. Using GPT-4 for carbon estimation, the game simulates creating a dish where each ingredient adds to the total footprint. If the carbon limit is exceeded, the dish "explodes," teaching users the impact of their choices. The project also offers a public API for calculating carbon footprints, making it a versatile tool for both education and analysis.
- [June 2024]
- 🏆 Beated GPT-4 with Open Source. Thanks to the work I began with my benchmark and the subsequent collaboration with the Outlines community, we successfully outperformed the best-performing GPT-4 model at a task that OpenAI essentially invented: function calling.
- 😂 Released Datarisas, a database to track the history of chilean humor.
- [May 2024] 💬 Released Chatstract, a tool for progressively extracts structured information from multi-turn user chats using pydantic schemas and LLMs. Powered by OpenAI API + Instructor.
- [April 2024]
- 📚 Released the Repository Template for Course Websites, a tool for educators to easily create and manage course sites using GitHub Pages, powered by GitHub Actions and Material for MkDocs. I realized I was repeating this process every time I created a course, so why not automate it a bit?
- 🚀 Launched a GitHub project to scrape RSS feeds into daily json files, driven by the need for a quick dataset to grasp "what's happening in the country." Automated with GitHub Actions, it runs every 24 hours. This is part of a larger project—stay tuned.
- [February 2024] 💡 I'm currently working with Subconscious.ai to help build an AI Copilot for Causal Market Research. It's my first time working with Langchain's LCEL, and I've managed to shift the whole query process for LLMs into this new framework.
- [January 2024] 🌌 I recently taught an intensive data science workshop titled "Orbiting Jupiter and Landing on Earth: A Journey in GitHub Towards Practical and Collaborative Data Science". This course focused on transforming the way students approach data science, shifting from solitary projects in Jupyter Notebook to a more collaborative and practical approach, ready for integration into real software environments.
- [December 2023] 📖 Co-created the AI engine for the "Explorador Constitucional" a joint project by La Tercera (one of Chile's largest newspapers) and CRiSSLab. Deployed an OpenAI API-based product in production, which efficiently searches, summarizes, and compares constitutional texts. The site attracted over 5,000 users per day. The first version of the code is here.