AI Proof of Concept : 8-Week Engagement

The AltaML AI PoC engagement is typically 8 -12 week long, and includes ideation, problem statement/ML model canvas creation, followed by a feasibility assessment, data exploration and then AI/ML model experimentation leveraging the Azure AI stack. The PoC engagement then validates the models with new data, assessing business impact and workflow integration. This approach efficiently de-risks the ML process, hastens ROI realisation, and supports informed decisions for full-scale deployment.

Phase 1: Feasibility

Key Activities:

Discovery and Exploration: Validate AltaML’s understanding of the business problem, gain access to data, define success criteria, estimate ROI, and propose a technical work plan.

Data Assessment: Conduct exploratory data analysis (EDA) to understand the context of data sets, important features, and provide an assessment of overall data quality.

Machine Learning Approaches Assessment: Identify relevant techniques, suitable Azure AI tools, open-source/pre-trained models and begin light experimentation to inform the PoC models/techniques/Azure AI tools leveraged.

Deliverables: Feasibility Assessment Report:

• Use case definition and description
• Defined problem statement
• Project assumptions and potential risks
• Data source(s), including an analysis on data sufficiency, quality and complexity
• Responsible AI for ethics and bias
• Potential machine learning techniques and Azure AI tools and frameworks
• Draft infrastructure architecture

Phase 2: Proof of Concept (PoC)

During this phase of the engagement, AltaML will conduct machine learning modelling, explore pre-trained models and Azure AI tools, and develop a PoC model. This PoC will be focused on demonstrating the viability of using Azure AI-powered machine learning for the selected use case. The outputs of the models will be evaluated based on the success criteria defined in Phase 1, and recommendations for next steps will be provided.

Key Activities:

Machine Learning Modeling: Use a combination of Azure AI tools, pre-trained models, fine tuning, and custom development to build the most accurate model. The model with the best performance will be selected.

Feature Engineering: Select, modify, and weight raw data features to enhance the performance of the machine learning model.

Model Evaluation: Evaluate model performance based on the pre-defined success criteria.

Model Improvements: Refine models to enhance performance.

Responsible AI: Assess the models ethical and bias risks.

Deliverables:

• PoC code
• PoC model demonstration
• Final presentation
• Pilot Proposal