February 24-26, 2027
Montreal, Canada
No Black Box: The Open Source Answer to AI App Builders
AI can build applications faster than ever, but what happens when you need to stay in control of your data, infrastructure, permissions and application logic?
In this 2-day hands-on workshop, we will explore how to integrate AI into real applications without relying on a black-box platform. Using an open source application platform as our case study, we will build and deploy an application, connect AI assistants to application data and tools, and examine what happens under the hood.
We will explore practical approaches to AI governance, permissions, tool design, error handling and human-in-the-loop workflows, as well as techniques for testing and improving the reliability of AI integrations.
You'll work through practical exercises using your own laptop and leave with a running application, along with patterns you can apply to safely integrate AI into applications and infrastructure you control.
In this 2-day hands-on workshop, we will explore how to integrate AI into real applications without relying on a black-box platform. Using an open source application platform as our case study, we will build and deploy an application, connect AI assistants to application data and tools, and examine what happens under the hood.
We will explore practical approaches to AI governance, permissions, tool design, error handling and human-in-the-loop workflows, as well as techniques for testing and improving the reliability of AI integrations.
You'll work through practical exercises using your own laptop and leave with a running application, along with patterns you can apply to safely integrate AI into applications and infrastructure you control.
Intro - AI Without the Black Box
- The trade-offs of AI application builders
- Keeping control of your data, infrastructure and application logic
- Open source and self-hosted approaches
- Understanding where AI fits into an application architecture
- Commercial versus locally hosted AI models
Part 1 - Building an Application with AI
- Understanding configuration-driven applications
- Building forms, connected data and reports
- Defining users, groups and permissions
- Connecting an AI assistant to an application
- Building applications from prompts and existing documents
- Modifying by hand what the AI creates
- Publishing and deploying the application
- Configuration as code and version control
Part 2 - Connecting AI to Applications
- Giving AI controlled access to application data and functionality
- Understanding tools, resources and prompts
- Connecting external AI clients through MCP
- Embedding an AI assistant directly into an application
- Designing which tools and capabilities should be exposed
- Working with commercial and locally hosted models
Part 3 - AI Governance and Permissions
- Using existing application permissions to control AI access
- Restricting what data an AI assistant can read
- Protecting sensitive and personally identifiable information
- Separating user and administrator capabilities
- Creating audit trails for AI actions
- Designing safe workflows for destructive operations
- Understanding prompt injection and other risks
Part 4 - Domain Knowledge and Human-in-the-Loop Workflows
- Giving AI the business context it needs to make useful decisions
- Keeping domain knowledge inside the application instead of individual prompts
- Designing rules that AI can interpret consistently
- Creating AI-assisted business processes
- Building work queues for recurring AI tasks
- Flag, act, review and approve workflows
- Handling incorrect or ambiguous AI decisions
Part 5 - Designing Tools AI Can Use Reliably
- Designing effective tool interfaces for AI
- Writing tool descriptions as instructions for models
- Choosing the right tool granularity
- Designing errors that help models recover
- Returning useful results after write operations
- Defining valid values and constraints
- Designing safe protocols for destructive operations
- Structuring tool responses to improve model accuracy
Part 6 - Testing and Improving AI Reliability
- Understanding context budgets and large tool surfaces
- Reducing and selecting tools to improve reliability
- Creating repeatable tests for AI integrations
- Measuring whether models select the correct tools and arguments
- Comparing behaviour across different models
- Testing failure scenarios and prompt injection
- Identifying and debugging unreliable AI behaviour
- Evaluating commercial versus local open-weight models
Participant should have:
- Intermediate skills
- Comfort with web application concepts
- Basic knowledge of HTML and SQL
- Basic understanding of permissions and API keys
- No prior experience with the case-study platform is required
Duration:
- 2 days
- 9:00 am to 5:00 pm
- 1 hour lunch break included at the hotel's restaurant
- 15 min coffee break every morning and afternoon
Julian Egelstaff
Formulize
Julian leads the Formulize open source project, a data management platform that turns messy information and chaotic processes into beautiful databases that blend smoothly with how people actually work. In 1996, Julian got a Bachelor of Journalism and Philosophy, and naturally went on to 30 years in the software industry. In 2003, he co-founded Freeform Solutions, a non-profit web development firm, where the Formulize project began. Somehow, along the way, Julian became a Zend Certified Engineer.
