Beyond Defaults: Smarter AI Integration

At the Indiana CIO Network’s AI Integration Roundtable, Calvin Hendryx-Parker, CTO of Six Feet Up and AWS Hero, shared a playbook for AI adoption. His core message: organizations cannot rely on defaults. Choosing the right models, grounding outputs in data, and scaling with discipline are essential to reducing risk and delivering value.

Pick Models With Intent

Too many enterprises still treat ChatGPT or Claude as plug-and-play tools. Calvin cautioned that these choices directly affect how data is handled, how reliable the outputs are, and whether systems can scale effectively.

“Evaluate models deliberately rather than relying on defaults,” said Calvin.

Calvin pointed to benchmarking resources like Hugging Face’s trust and performance leaderboards as a way to align model selection with enterprise priorities such as privacy, bias, and compliance. He also urged leaders to recognize the difference between true open source (which includes weights, data, and code) and “open weights” releases that offer only partial transparency.

Fine-tuning, Calvin added, should be used sparingly. It is best reserved for clear, high-value use cases such as image classification rather than treated as a default strategy.

Use RAG to Improve Accuracy and Privacy

Calvin pointed to retrieval-augmented generation (RAG) as a way to improve accuracy, protect sensitive data, and deliver citeable, auditable answers. By storing documents in vector databases and retrieving them at query time, teams can reduce hallucinations and keep private knowledge safe from third-party APIs. Read our blog for more detail on “How to Make AI Tell the Truth.”

To illustrate, Calvin ran a demo. He scraped a conference website, enriched the speaker bios, and built a chatbot that could answer questions the site itself could not, such as which talks covered AI or who to network with. The system ran entirely on his laptop, showing how organizations can deploy accurate, private, and cost-conscious AI.

Apply Software Discipline to AI Deployment

Calvin reminded the group that deploying AI is still a software engineering exercise. The same principles that govern applications such as planning, testing, monitoring, and optimization apply here as well.

Best Practices for Scaling AI

• De-risk operations: Add guardrails and frameworks to prevent bias, leakage, and hallucinations.
• Scale with visibility: Build observability from day one to track performance and manage costs.
• Accelerate delivery: Apply agentic AI to repetitive tasks, such as compliance controls, to cut delivery times from hours to minutes.
• Protect sensitive data: Run sensitive workloads on local or private-hosted models when cloud risks are unacceptable.
• Optimize performance: Push teams beyond defaults by tuning embeddings, chunking strategies, and frameworks to fit the workload.

Watch the Presentation

Curious how RAG, embeddings, and open source tools can transform AI from a risky experiment into a strategic advantage? Watch the full AI Integration Roundtable session.

Did Calvin’s session spark ideas for your AI strategy? Six Feet Up helps organizations modernize pipelines and adopt AI responsibly. Let’s talk.