Wireless Earbuds
Deploy natural, human-sounding audio with ElevenLabs. Create realistic, dynamic, and emotionally expressive voices for any project, from interactive AI companions to narrated stories and voice-enabled apps. ElevenLabs will empower you to build rich, immersive experiences without the need for actors or complex audio production, using simply the power of AI.

Integrate fully autonomous audio experiences into your hack with ElevenLabs and give your project a voice, along with giving your team the chance to win some wireless earbuds!

MLH Swag Kits
It’s time to push the boundaries of what’s possible with AI using Google Gemini. Check out the Gemini API to build AI-powered apps that make your friends say WHOA. So, what can Gemini do for your hackathon project?

Understand language like a human and build a chatbot that gives personalized advice
Analyze info like a supercomputer and create an app that summarizes complex research papers
Generate creative content like code, scripts, music, and more
Think of the possibilities… what will you build with the Google Gemini API this weekend?

Assorted Prizes
The world of development is evolving fast and Solana is leading the charge with a network built to handle all of your infrastructure needs. Forget high fees and slow confirmations, it’s time to build applications that are fast, efficient, and scalable.

Harness Solana’s core advantages like blazing fast execution and near-zero transaction costs to make your hackathon ideas become real world projects. With Solana, the possibilities are endless.

Create a game, social app, or consumer product that relies on instant, high-frequency transactions.
Design a sophisticated trading, lending, or decentralized exchange (DEX).
Build a prototype for supply chain, identity, or payments that can handle massive, real-world volume.
Show us how you can innovate with Solana for a chance to win some cool prizes for you and each member of your team!

Stream Deck Mini
Building high-performance apps under hackathon deadlines shouldn't mean spending half your weekend configuring complex databases or learning obscure query languages. Tiger Data extends PostgreSQL to give student developers an ultra-fast foundation for real-time data, time-series metrics, and complex analytics.

Why build with Tiger Data this weekend?

Standard SQL Power: Query massive datasets instantly using the SQL you already know, with no complex NoSQL aggregation pipelines required.

Unified Data Stack: Store relational user profiles and high-frequency metric streams side-by-side in a single database.

Instant Dashboards: Serve lag-free, real-time frontend charts using pre-computed Continuous Aggregates.

90%+ Data Compression: Store millions of events on free-tier cloud instances without running out of disk space.

Whether you’re building real-time IoT monitoring, AI-driven analytics dashboards, or financial prediction engines, show us how Tiger Data powers your project. The team that demonstrates the most innovative, impactful, and performance-driven use of Tiger Data will take home some great prizes!

Tile Essentials Pack
Every AI model API is stateless by default. That means your app forgets everything the second a session ends. State management should be the first step of any AI build, regardless of what LLM you are using, and Backboard gives you that for free. But Backboard is not just state management. It is a single, unified API built on the world’s #1 AI memory that gives you everything you need in one place: long-term memory, RAG, embeddings, tool calls, model routing across 17,000+ LLMs, and persistent context that stays alive across every page refresh, session, and user. No stitching together five different services. One API. One integration. Built on the best memory in AI.

Here’s just a small list of things Backboard can make possible for your AI app!

AI-powered Travel Guide: Remembers allergies and preferences from past trips.
Personalized Fitness Coach: Adjusts workouts based on progress and injury history.
Smart Home Controller: Learns routines over time to anticipate lighting and climate preferences!

Use Backboard to build a seamless, persistent user experience without the usual infrastructure headache! Each winning team member will receive a Tiles Essentials Pack because, just like AI, we all need an occasional reminder of where things are!

M5Stack Official Tab5
Vultr empowers hackers to bring their high-performance projects to life instantly; providing everything from the speed of one-click deployment and scalable cloud compute, to specialized Vultr Cloud GPUs that can power AI-driven applications. We want you to push the limits of what can be built when infrastructure is no longer the bottleneck!

Sign up for a Vultr account today and claim your free cloud credits! Take your next hack to the cloud with Vultr for a chance to win some awesome prizes for you and your team!

What Should Happen Next?
Las organizaciones generan miles de señales todos los días: interacciones, eventos, mensajes, cambios de estado y comportamiento de usuarios.
El reto consiste en desarrollar un agente de inteligencia artificial capaz de interpretar estas señales y determinar cuál debería ser la siguiente mejor acción.
La solución deberá:
- procesar eventos;
- entender contexto;
- detectar intención;
- estimar riesgo;
- recomendar una acción o innaccion;
- justificarla;
- determinar el momento adecuado;
- indicar su nivel de confianza.

Asistente virtual de compras y soporte
PROBLEMÁTICA
Los clientes preguntan lo mismo una y otra vez: estatus de pedido, devoluciones, garantías, disponibilidad. El soporte se satura, las respuestas tardan y la compra se enfría.
OBJETIVO
Diseñar un chatbot que responda con base en el catálogo y las políticas de una tienda, y sepa escalar a una persona cuando no tenga certeza.
ALCANCE
Prueba de concepto con catálogo y políticas ficticias, y una interfaz web de chat funcional.
HERRAMIENTAS GRATUITAS Y DATOS
Python, Hugging Face, Ollama (modelos abiertos), LangChain o Llamalndex y Gradio; o niveles gratuitos de APls de LLM.
Datos: catálogo y FAQ simulados.

Almacén inteligente para taller
El desafío consiste en diseñar e implementar un servicio de gestión inteligente de inventario para un taller especializado en la reparación de equipo industrial, particularmente husillos y motores eléctricos.
La solución deberá integrarse con el ecosistema operativo existente mediante una arquitectura orientada a eventos, procesando en tiempo real la información generada por los distintos procesos del taller y manteniendo un control consistente de existencias, reservas, faltantes y movimientos de inventario.

ANTES DE QUE SUENE EL TELEFONO
la idea es simple
una pyme tiene un problema de ti y se entera cuando el cliente llama enojado
su mision es detectarlo antes de esa llamada
ustedes son el sistema de alerta temprana
mientras mas antes avisen de un incidente real mejor
pero si avisan por todo nadie les va a creer
QUE RECIBEN
datos sinteticos de 4 empresas
gracko es un despacho contable
gramo
kaiku es una tienda en linea
ttr es de logistica
de cada una tienen 3 archivos
contexto.md dice que hace la empresa y sus horarios y eventos
metricas.csv trae una fila por minuto con cpu ram disco latencia_ms y errores
logs.md trae logs de servidor firewall y app
ademas todos_metricas.csv junta las 4 empresas con una columna tenant
hay 54 corridas distintas en data/corridas
usenlas para entrenar validar y probar
repo https://github.com/brunovillarrrrrr/zikit-dataset
QUE TIENEN QUE HACER
entregar una lista de alertas
cada alerta llevan tenant timestamp y confianza
opcional el tipo de incidente que creen que viene
LO DIFICIL
cada empresa tiene 3 incidentes reales por corrida
algunos avisan con 40 minutos y otros con mas de un dia
hay señuelos como respaldos nocturnos cierres de nomina y cargas masivas
se ven feos pero no son incidentes
si alertan por ellos cuenta como falsa alarma
los datos tienen ruido celdas vacias y apagones
los logs traen errores que no significa nada
las pistas reales son pocas
el contexto del negocio ayuda a saber que es normal
REGLAS
solo usen los datos del kit
no se vale mirar el futuro
una alerta en el minuto t solo usa datos hasta t
las carpetas _solucion son del evaluador
no las usen para entrenar ni para predecir
COMO SE CALIFICA
incidentes reales detectados antes del impacto
pocas falsas alarmas
cuanto tiempos de anticipacion ganaron
mas temprano es mejor