Scaling beyond PoCs

Artificial Intelligence (AI) Strategy

Vidhi Chugh

AI strategist and ethicist

What does it mean to scale?

 

 

• Scalability:
  • Manage increasing data, users and tasks
  • Maintaining performance, quality and affordability

 

• Scaling AI = 3x higher returns!
  • Integrate into existing business processes

Cloud computing platforms

 

• PoC to full-scale development

 

• Cloud computing platform
  • Compliance with data regulations
  • Security
  • Support multi-cloud and hybrid-cloud architectures

 

Deployment considerations

• Cost and performance analysis
• Governance measures; ease of deployment and management

Build vs. buy

 

• Off-the-shelf model vs. custom algorithm

• Greater control over system performance

 

• Project timelines

• Custom algorithm takes time as compared to an available algorithm

Scalability and performance

 

• Scaling = quality predictions without compromising performance

 

• Performance:
  • Accuracy
  • Latency: time model takes for prediction
  • Throughput: Number of predictions in a given time

 

 

Data at scale

• Core of machine learning models

• Managing data lifecycle

  • Preparing and maintaining

Access to data

 

• Accessible to specific departments; isolated from rest
  • Results in silos
  • Reduces transparency
  • Inefficient models

 

• Different sources = different formats
  • Standardize

 

Architecture that can scale

 

 

• Designing the right architecture

 

• Microservices
  • Scale independently
  • Flexibility
  • Responsiveness

 

• Coordinating interaction is difficult

Infrastructure to scale

 

• Extensive data computations
• GPUs speed up tasks but at increased cost

 

• Weigh the benefits
  • Expedite model training
  • Improved performance
  • At additional costs

Streamline AI processes

• Detailed implementation plan and architecture design
• Standardize and streamline
  • Model building, deployment, and monitoring

Let's practice!

Artificial Intelligence (AI) Strategy