AI Resources

Everything the four Saturdays rely on, in one place, for Ayiti AI Hackathon participants. AI coding tools, LLM APIs, agent and RAG building blocks, Kreyòl speech, messaging and MonCash integration, on-device inference, and deployment — the tools you will actually use between now and November 29.

Whether you are building a tool-calling agent, a model that runs on a phone, or an assistant that speaks Kreyòl, start here. Free tiers and partner credits cover everything in the program.

Essential Tools

Learning Resources

AI Coding Tools

Cursor DocumentationClaude Code DocumentationGitHub Copilot Docs

Agents, MCP & RAG

LangGraph DocumentationVercel AI SDK DocumentationClaude Tool Use GuideMCP Documentation

Open Models & Local Inference

Hugging Face Transformers Docsllama.cpp (Local & Edge Inference)MLC LLM (Android & iOS On-Device)

Hands-On Notebooks

Unsloth Fine-Tuning NotebooksAnthropic Cookbook (Agents & Tool Use)Makandal Training NotebookDatasets Preparation Guide

Haitian Creole Models & Speech

Haitian Creole Models (Makandal)Makandal Pre-trained ModelHaitian Creole Datasets

RAG Quality & Evaluation

RAG Tutorial (LangChain)Vector Search with pgvector (Supabase)Hugging Face Evaluate

Specialized Resources

Deployment

VercelFly.ioHugging Face Spaces

Kreyòl Speech & Data

Haitian Creole Text Data (Makandal corpus)Meta MMS — Speech Recognition for 1,000+ Languages (incl. Haitian Creole)Whisper — Open Speech-to-Text

Deployment Platforms

Gradio DocumentationDocker (Containerize Your Prototype)Hugging Face Inference Providers

RAG Quality & Evaluation

Ragas — RAG Evaluation FrameworkEmbeddings Guide (Hugging Face)Hugging Face Evaluate

Key Frameworks & Libraries

  • Claude Code: Agentic coding tool used throughout the program to build, test, and ship faster

  • LangGraph: Stateful agent workflows with tool calling, memory, and human-in-the-loop control

  • Vercel AI SDK: Provider-agnostic TypeScript toolkit for LLM apps, agents, and streaming UIs

  • Model Context Protocol: Open protocol for connecting agents to external data and tools

  • Transformers: State-of-the-art open models for text, vision, audio, and multimodal tasks

  • llama.cpp: Run quantized models on CPUs, phones, and single-board computers — the engine behind local inference

  • Unsloth: Fast, memory-efficient fine-tuning of open models, including small Kreyòl models

  • Gradio: Build ML demos and voice or chat interfaces with minimal code

  • Datasets: Load, clean, and share text, audio, and vision datasets — including Haitian Creole corpora

  • Evaluate & Ragas: Reproducible metrics for models and RAG pipelines: accuracy, relevance, faithfulness, word error rate

Getting Started Before Saturday 1

Step 1

Set Up Your Tools

Install Claude Code or Cursor, create a GitHub account, and get API keys for Claude, OpenAI, or Gemini; partner credits are distributed at the first session

Step 2

Pick Your Track

A — tool-calling agent, B — local or on-device model, C — Kreyòl voice; choose with your team according to your frontier and keep it for all four Saturdays

Step 3

Build the Core

Ship a first prototype on Saturday 1, add the agent layer on Saturday 2, ground it in data on Saturday 3

Step 4

Make It Resilient

Design for failure, pass the unplug test, and rehearse the five-minute demo on Saturday 4

~~

Ayiti AI

Ayiti AI Hackathon 2026 | AI Beyond Ideal Conditions