In this conversation, Tommy Levy, Senior Director of AI at Agiloft, discusses the challenges and processes involved in building AI and ML features for contract lifecycle management. He talks about the work Agiloft is doing to extract key terms and labels from contracts using natural language processing. He also explains their redlining feature, which blends existing language with standard language and summarizes changes. Tommy emphasizes the importance of quantifying risk and uncertainty in ML development and adapting agile methodologies. He also highlights the unique considerations in the SDLC for data science teams, such as data lineage and data provenance. In this conversation, Tommy discusses the challenges of integrating AI/ML features into the software development life cycle (SDLC) and the importance of tracking data along with the model. He mentions the existing tooling and platforms for MLOps but highlights the need for a more comprehensive solution. Tommy also explains the team structure at EHLoft, with data science teams focusing on R&D, an MLOps team building infrastructure and tooling, and a Legal Knowledge Engineering team providing domain expertise. He shares the ongoing initiatives to standardize the offline and online experiences and prioritize based on business value and opportunity cost.

‍

Takeaways

Agiloft is using AI and ML to extract key terms and labels from contracts for contract lifecycle management.

Their redlining feature blends existing language with standard language and summarizes changes.Quantifying risk and uncertainty is crucial in ML development, and agile methodologies can be adapted for ML projects.

Data lineage and data provenance are important considerations in the SDLC for data science teams. Integrating AI/ML features into the SDLC requires tracking data along with the model at every step.

Existing tooling and platforms for MLOps provide some support, but a comprehensive solution is still needed.* The team is working on standardizing the offline and online experiences to streamline development and deployment.

Prioritizing based on business value and opportunity cost is crucial for maximizing ROI.

Highlights:

🎯 Prioritize tasks that deliver business value

🔄 Constantly assess where you are spending time and energy

💡 Uncertainty in AI requires research and discovery phase

🚀 Agile methodology can be applied to machine learning developmentKey Insights

💼 Understanding opportunity cost helps in making informed decisions about resource allocation and prioritization.

🧪 Research and discovery phase is crucial in AI development to assess feasibility and value of a feature or model.

🔄 Agile methodologies can be applied to machine learning development by breaking down work into sprints and continuously iterating based on feedback.📊 Building standard tooling and infrastructure streamlines the development process and improves efficiency.

🌟 Collaboration between data scientists, ML engineers, and domain experts is essential to ensure effective communication and alignment of goals.

📈 Regular retrospectives and feedback loops help in identifying areas of improvement and enhancing the development process.

🌍 Consideration of business viability, including legal and privacy concerns, is necessary for successful implementation of AI features.

Chapters:

00:00 - Introduction and Background

03:01 - Challenges in Contract Lifecycle Management

07:40 - Dealing with Non-Deterministic Nature of Software

10:04 - Applying Software Engineering Principles to AI/ML Engineering

11:08 - Software Development Life Cycle for Data Science Teams

22:25 - The Integration of Data and Models25:10 - Team Structure at AgiLoft

29:13 - MLOps and Developer Experience

33:44 - Improving the Developer Experience

40:36 - Maximizing ROI and Prioritization