Por qué los LLMs le fallan a tu tía — y el middleware que lo arregla

Learn how to build robust LLM agents for elderly users with low digital literacy. This talk details architectural fixes for error handling, vocabulary adaptation, and LLM routing, demonstrated with real user sessions.

Python python-telegram-bot Claude Opus 4 Gemini Flash n8n

Overview

Un agente en OpenClaw/Python para adultos mayores con baja literacidad digital, accesible por WhatsApp/Telegram. Lo presenté en el hackathon de Tribu IA el 13 de abril con fallas graves. Esta demo muestra el diagnóstico, las tres correcciones de arquitectura, y los resultados de una sesión de prueba real con dos usuarias de +85 años.

Tech stack

Python

Python: The high-level, general-purpose language built for readability, powering everything from web backends to advanced machine learning models.

Python is the high-level, general-purpose language prioritizing clear, readable syntax (via significant indentation), ensuring rapid development for any team . Its ecosystem is massive: use it for robust web development with frameworks like Django and Flask, or leverage its power in data science with libraries such as Pandas and NumPy . The Python Package Index (PyPI) provides thousands of community-contributed modules, offering immediate solutions for tasks from network programming to GUI creation . The language is actively maintained by the Python Software Foundation (PSF), with the stable release currently at Python 3.14.0 (as of November 2025) .

https://python.org

View projects
python-telegram-bot

A pure Python, asynchronous interface for the Telegram Bot API, utilizing the `telegram.ext` module for high-level bot development.

This library delivers a pure Python, asynchronous interface for the official Telegram Bot API, compatible with Python 3.9+ environments. It goes beyond the raw API with high-level classes: specifically, the `telegram.ext` submodule provides the `Application` and various `Handler` classes (e.g., `CommandHandler`, `MessageHandler`) to streamline complex bot logic, message routing, and conversation flows. Developers use this structure to efficiently manage updates and build robust, scalable Telegram bots with minimal boilerplate code.

https://python-telegram-bot.org

View projects
Claude Opus 4

Claude 4 Opus is Anthropic’s next-generation frontier model designed for high-stakes reasoning and complex multi-step orchestration.

Claude 4 Opus represents the peak of the Claude 4 family, delivering significant gains in coding proficiency and nuanced linguistic understanding. It handles massive 200k-token contexts with near-perfect recall (99% accuracy) and excels at autonomous task execution. Built on updated Constitutional AI frameworks, Opus 4 provides enterprise-grade reliability for quantitative analysis and creative synthesis while maintaining a near-zero hallucination rate in technical documentation.

https://www.anthropic.com/claude

View projects
Gemini Flash

Google's most balanced and efficient multimodal model: superior speed, cost-effectiveness, and a 1 million token context window for high-frequency tasks.

Gemini 2.5 Flash is our high-performance, cost-optimized model, engineered for speed and efficiency in agentic workflows. As a natively multimodal model, it processes inputs including text, code, images, audio, and video, available to developers via the Gemini API and Vertex AI . This model delivers a massive 1,048,576 input token limit and introduces 'thinking' capabilities for enhanced reasoning and accuracy in complex tasks like summarization and data extraction . It provides a superior balance of quality and low latency, outperforming previous versions like Gemini 1.5 Pro on key benchmarks while operating at twice the speed .

https://ai.google.dev/models/gemini

View projects
n8n

n8n (pronounced 'n-eight-n') is the fair-code workflow automation platform: it gives technical teams the power of custom code (JavaScript/Python) via a visual, node-based editor.

This is n8n: a powerful, self-hosted workflow automation platform designed for developers and technical teams. It uniquely blends the speed of a no-code UI with the flexibility of custom code, allowing users to build complex, multi-step automations that other tools cannot handle. With 400+ integrations and native AI capabilities, n8n ensures you maintain full control over your data and deployment (on-prem or cloud). For example, you can build a workflow to automatically pull data from a private API, transform it with a custom Python script, and push the results to both Salesforce and a Slack channel, all from a single visual canvas.

https://n8n.io/

View projects

Por qué los LLMs le fallan a tu tía — y el middleware que lo arregla

Related talks