4 Saturdays
Three-hour live sessions, 12:30–3:30 PM (Port-au-Prince), October 31 to November 21
Four Saturdays to get ready. The Ayiti AI training is free, online, live, and recorded. It is mandatory for the thirty selected teams and open to the whole community as auditors. Teams that prepare seriously are the teams that ship. In 2025, that was the single clearest lesson.
Each team picks one track in the first session and builds the same prototype through all four. By November 27, every team arrives with a working application grounded in a real user, tested against a real failure, and a 48-hour plan a mentor has already reviewed.
4 Saturdays
Three-hour live sessions, 12:30–3:30 PM (Port-au-Prince), October 31 to November 21
Live on Zoom
Interactive classes, recordings the same evening, and daily support on Discord
Build Every Week
One track per team, one deliverable due each Friday, and API credits from our partners distributed at the first session
Build and deploy an AI application with LLM APIs and AI coding tools, from idea to working prototype
Design an agent that calls tools, keeps memory, handles errors, and does real work through WhatsApp, SMS, or MonCash
Run a model locally or on-device — a laptop, an Android phone, a Raspberry Pi — without depending on the cloud
Integrate Kreyòl speech recognition and synthesis into a product people can talk to
Ground an application in your own data with RAG, evaluate it, and make it work offline
Design for failure, pass an unplug test, and deliver a five-minute demo that convinces
Ship Your First AI Application
LLM APIs, effective prompting, and AI coding tools (Claude Code, Cursor, GitHub Copilot). Move from an idea to a deployed prototype for real users. Pick your track: A — a tool-calling agent, B — a quantized model running locally, or C — a Kreyòl voice assistant. Deliverable: a working prototype and a user brief naming one user, one problem, and one real-world constraint.
Build AI Agents That Do Real Work
Tool calling, memory, Model Context Protocol (MCP), workflow orchestration, and error handling. Connect your agent to real services: A — WhatsApp or SMS with the MonCash sandbox, B — on-device inference on Android or Raspberry Pi, C — Kreyòl speech in the agent loop. Deliverables: a functioning agent, three logged user conversations, and your user brief.
Build AI Applications with Your Own Data
Retrieval-Augmented Generation, document processing, embeddings, vector search, data cleaning, synthetic data, and basic evaluation. A — business-document RAG with source references, B — offline-first RAG that syncs when connectivity returns, C — a Haitian Creole text or speech data pipeline. Deliverables: a data plan, a working core component, and first measurable evaluation results.
Build Resilient AI Systems for the Real World
Graceful degradation, fallback ladders (cloud → SMS → offline), recovery from power and network failure, and scaling beyond Haiti. Lab leads run an unplug test on every prototype. Deliverables: a mentor-reviewed 48-hour plan, a five-minute demo script, and a passed unplug test.
Weekly Deliverables
Saturday 1: Working prototype and user brief
Saturday 2: Functioning agent and three logged conversations
Saturday 3: Data plan, core component, and first evaluation results
Saturday 4: Mentor-reviewed 48-hour plan, demo script, and unplug test
Between Sessions
Live demos and peer reviews every Saturday
Project follow-up during the week, announced to each cohort
Discord support, every day
Prepare like it matters. It does.
Two unjustified absences from the four Saturdays and your seat goes to the waitlist. That rule protects the thirty teams who come to build. Everyone else is welcome to follow along as an auditor.
Saturdays, October 31 – November 21, 2026 · 12:30–3:30 PM · Live on Zoom
Ayiti AI