Using an AI-Driven Model for Self-Service API Integration Developers must often build new applications that pull together multiple back-end services delivered via APIs. A new mobile app, for example, may need to connect to backend ERP or CRM systems using a variety of third-party APIs, which are then integrated and delivered as a new API that provides the mobile app interface. By applying machine learning to this process, developers can greatly increase the speed and reduce the complexity of API integration and management, empowering non-developers with the benefits of API integration.
This presentation outlines the challenges of API integration in a modern cloud context and explains how developers can leverage machine learning to speed application development, reduce errors and improve security and compliance.
9. API based integration architecture
Public APIPartner APIPrivate API
Internal Channels API Consumer
API Platform
System of Insight
Cloud
Integration
B2B
Integration
Engagement APIs
(Microservices)
Data,ComputeGrid
API Provider (SaaS, Partner, Customer)
System of Records
Integration (ESB/ETL/MFT/ System APIs)
iPaaS
Process Orchestration Layer
Engagement Processes (Process / Functional APIs)
10. 10
Connecting to the API
Integrating with the API
Authenticate
Error Handling
Events & Polling
Workflow
Orchestration
Bulk
Custom
Objects
Versions
Lifecycle
Learning
Map &
Transform
Testing
13. 13
Marketing
Dev QA Sup
Finance Sales
Dev QA Sup Dev QA Sup
INTEGRATION PLATFORM SERVICES
Self Service AutomationGovernanceDesign
Pros: ZERO Lead Time
Faster Integration Dev : 1x
LOB is autonomous
Cons: Guardrails for establishing standards
Marketing Finance Sales
CENTRALIZED INTEGRATION SERVICES
SupportQADev
Pros: Integration Domain Expertise
Maintain Integration Standards
Cons: Longer Lead Time
Longer Dev Time : 6x
Priority Alignment Issues
Centralized model Self service
14. User experience
● Abstraction of API complexity
● Patterns
● Security(certification/authorization)
● No Management of Infrastructure
● Low code to enable non-developers
23. API recommendations with ML
• A recommendation model can be
used to provide good guesses for
API parameters and common API
sequences.
• API usage and interaction patterns
can be learned from examples.
Learning new APIs is time consuming.
Learning conventions and API
interactions is more time consuming.
!
24. Potential for API recommendations
Weather Underground (Weather Observation)
◦ "observation_time_rfc822": "Wed, 27 Jun 2018 17:27:13 -0700",
Twitter (Tweet)
◦ "created_at": ”Wed Jun 27 17:27:13 +0000 2018"
GitHub (Repository)
◦ "created_at": "2018-06-27 T17:27:13Z",
25. API mapping and integration
• required fields and examples
• most commonly used fields
• mappings from input fields to destination fields
• Recommend API interaction patterns
Recommend
26. Machine learning stages
Data
Collection
Collect and prepare data
Data
Preparation
Make sense of data
ML Model
Training &
Testing
Use data to answer questions
Model
Deployment
Deploy and operationalize models
26
27. Data collection
● API documentation
● GitHub and other public repositories
● Platforms for API integration
31. Training and testing
Segment + user + org + project
Neural networks
Recommendation model
JSON
PARSER
FILE
READER
MAPPER
32. Model deployment architecture
User web app
Back-end
services
Storage
Storage
replica
ML trainingS3 file systemML APIs
MachineLearning
Metadata
Analytics
34. Lessons learned
● Need lots of Examples
● Data prep
● Choosing and adjusting the right model
● Trial and Error
● Iterative improvement
● ML as API
● AI enables better API’s and API’s enables better AI
34
35. Applications
● Competitive edge through best of breed applications
● Digital transformation that scales through self-service
● M&A activity integrating different groups with varied skillsets and apps through
API’s
35
40. 40
Connecting to the API
Integrating with the API
Authenticate
Error Handling
Map &
Events & Polling Workflow
Orchestration
Bulk
Custom
Objects
Versions
Lifecycle
Learning
Transform
Testing