GeoAI · Maven live cohort

Building GeoAI Agents for Fun & Profit

Stack MCP, AG-UI, and A2A protocols week by week — from tool calling to orchestration and human-in-the-loop UI.

2-week global cohort · starts 13 July 2026Live sessions: Sat 18 Jul + Sat 25 Jul · US Central · 10:00 AM to 4:00 PM · recorded

Enroll on Maven View curriculum

Stack the architecture: Agent → + MCP → + AG-UI → + A2A → + HITLConversation for intent · map for results · specialists when you need them

Human-in-the-loop geospatial analysis

When a place is ambiguous, the app asks for clarification, you approve the analysis, and catchments appear on the map.

Deliverables

What you'll build

Each layer adds capability — Agent, then MCP, then AG-UI, then A2A — with runnable Colab or local code and a project that proves the stack works.

Week	You will…
Before start	Environment checkpoint — API keys, Exercise 0, Colab or local ready
Week 1 · MCP	MCP protocol — expose geospatial APIs as tools your agent can call (Exercise 1)
Week 1 · AG-UI	AG-UI protocol — connect browser chat to agent and map (Exercises 2–3) · project due Wed 22 Jul
Week 2 · MCP	MCP architecture — multi-tool servers behind one agent host (Exercise 5) · project due Fri 24 Jul
Week 2 · A2A	A2A protocol — orchestrator delegates to specialist agents (Exercise 6)
Capstone	Extend the orchestrator or document production deploy (MCP + A2A + AG-UI) · due Tue 28 Jul
Bonus	HITL protocol plus production patterns (Geobase skills, OSRM backend swap)

Calendar

Cohort schedule

Live sessions are recorded the same day. Async learners watch recordings, complete HOW lessons in order, and use office hours for unblock.

Sun 12 Jul

Student portal opens

Mon 13 Jul

Cohort starts · Week 0 + Week 1 released

Tue 14 Jul

Setup Q&A · 12:00 PM US Central / 7:00 PM Berlin (optional)

Wed 15 Jul

Project: Environment checkpoint (due)

Sat 18 Jul

Live 1 · 10:00 AM to 4:00 PM US Central (Exercises 1 to 3, recorded)

Wed 22 Jul

Office hours + Project: MCP + AG-UI integration (due)

Fri 24 Jul

Project: Multi-tool MCP + AG-UI (due)

Sat 25 Jul

Live 2 · 10:00 AM to 4:00 PM US Central (Exercises 5 and 6, recorded)

Tue 28 Jul

Capstone: MCP + A2A + AG-UI (due)

Wed 29 Jul

Office hour: capstone Q&A · 12:00 PM US Central / 7:00 PM Berlin

Skills

Key outcomes

Every outcome is phrased as something you can do — not vocabulary to memorize.

Expose geospatial APIs as MCP tools your agent can call
Connect AG-UI so a browser chat hosts your agent session
Separate conversation from map results using client-side tools
Compose multiple MCP tools behind one server as requirements grow
Orchestrate specialist agents with A2A from a single AG-UI host
Add HITL and dynamic UI when the agent must ask a human first

Syllabus

Curriculum

Stack Agent → MCP → AG-UI → A2A across two live Saturdays. HOW lessons in Maven, recorded builds, and projects that mirror each layer.

Before you start

13–15 Jul

How the cohort runs (calendar, live vs recording)
How to add API keys (Colab Secrets + local .env)
Exercise 0 — verify Colab or local setup
Preview the stack: Agent → MCP → AG-UI → A2A
Project: Environment checkpoint (due Wed 15 Jul)

Week 1

Live Sat 18 Jul · US Central

MCP — tool calling for routing APIs (Exercise 1)
AG-UI — chat UI hosts the agent (Exercise 2)
MCP + AG-UI — client tools update the map (Exercise 3)
Live: I do / We do / You do on Exercises 1–3 (recorded)
Project: MCP + AG-UI integration (due Wed 22 Jul)

Week 2

Live Sat 25 Jul · US Central

MCP — multi-tool servers (Exercise 5)
A2A — orchestrator and specialist agents (Exercise 6)
Live: MCP composition lab + A2A lab + capstone kickoff (recorded)
Project: Multi-tool MCP + AG-UI (due Fri 24 Jul)
Capstone: MCP + A2A + AG-UI (due Tue 28 Jul)

Week 2 in action

Name two cities, get a route on the map, and pull in weather from a separate specialist agent.

Bonus

Unlocked Sat 25 Jul PM

HITL — MCP elicitation and dynamic UI (Exercise 7)
Production data plane — Geobase and agent skills (Exercise 8)
Alternate routing backend — OSRM with the same MCP + AG-UI pattern (Exercise 4)

From the workshop

Live teaching & hands-on labs

Photos from the GeoAI 2026 live workshop — same material, extended into a two-week Maven cohort with projects and office hours.

Shoaib Burq teaching the GeoAI workshop with a Tools slide on screen — Live session — how AI tools connect to maps and external services

Hands-on GeoAI lab with students coding on laptops — I do / We do / You do — students building routing tools in Python

Routing map chat application — Week 1 — ask for a route in chat, see it on the map

Geocoding and routing map chat — Week 2 — route between places you name in plain language

Human-in-the-loop geospatial analysis with isochrone catchments in chat — Bonus — keep humans in the loop for sensitive analysis

Social proof

What attendees say

Feedback from people who joined the live GeoAI workshop — hands-on MCP, map chat, and multi-agent orchestration.

“I came in knowing MCP was a buzzword and left with working tools I could call from an agent the same afternoon. The step-by-step labs beat any slide deck I've sat through.”

“Seeing the orchestrator hand off routing and weather to separate A2A agents felt like a pattern I could take back to work on Monday — not a toy demo.”

“Best geospatial AI session I've been to — real API keys, working labs you run yourself, honest trade-offs between MCP and A2A. No hand-waving.”

Instructor

Learn with Shoaib Burq

Shoaib Burq teaches agentic geospatial systems at Decision Labs — practical patterns for connecting AI assistants to maps, routing, and real-world location data.

The cohort follows the same GeoAI 2026 workshop stack — Colab notebooks, local Next.js labs, projects, office hours, and recorded Saturday lives — with materials on Maven after you enroll.

Stay in touch

Subscribe to the Decision Labs newsletter for GeoAI updates, new Colab exercises, and future cohort dates.

Newsletter

Ready to build?

Join the July 2026 cohort on Maven — starts 13 July 2026 · live Saturdays, async HOW lessons, and real map-and-chat projects.

Enroll on Maven