Presented by Contemporary Analysis & Omaha Data Science Academy​

CANalytics 2026 in the Press Thank you to Lev Gringauz and Silicon Prairie News for attending CANalytics (for the second year in a row) and writing this great article! In case you missed it, here are Lev's 5 key takeaways: Solve the right problems Know your audience Use GEPA to save money on AI tools Quality assurance is changing in the age of AI Don’t forget to test fairness in AI tools ​Read the article for a breakdown of each takeaway. ​

We're diving into #3 in this week's 2-minute lesson...

Become Highly Technical in 2 Minutes

This is an excerpt from “Inference management: What to do when your AI costs a fortune to run" presented by Tyler Hayes, Director of AI at CAN, and Ben Zwiener, Senior Data Engineer at CANalytics 2026.

We were promised models would get cheaper. They are not. In fact, they are getting exponentially more expensive, if you don't build cost-saving solutions.

One way to manage both cost and performance is through prompt optimization.

The Proof

Our team built an internal tool called Phosphor, an AI workflow that automatically pulls meeting transcripts and turns them into project management updates. Originally we ran Phosphor on Sonnet, it cost $200–$300 a month. We tried swapping to a smaller model (Qwen 27B) without changing anything else, performance dropped meaningfully.

We began using GEPA to optimize prompts on the small model, and performance didn't just recover, it exceeded the original frontier model. And cost dropped to $21 a month. That's over $2,000 saved per year.

And while GEPA might sound like the star of the show, your optimizer is only as helpful as your evaluation framework.

Understanding Evaluation

An eval has three parts: a dataset of hand-curated golden test cases and production examples you add over time; a metric function that scores outputs on a scale: not pass/fail, but how close or directionally correct the output was; and a process you can rerun anytime.

The eval is a durable asset. It grows over time and is what we fall back on to measure and understand performance. Over time it becomes the most accurate picture of how your AI is actually behaving.

What is an Optimizer?

An optimizer, such as GEPA, takes your eval and continually improves your prompts. It captures full traces of where your current prompt is failing, uses a larger model to diagnose why, proposes targeted mutations, tests them, and keeps the ones that score better.

Optimizers are commodities. New ones come out every six months, and you can swap between them. They're powerful. but only as good as what you're measuring them against. You cannot optimize what you cannot measure.

You want to invest your time in the eval, not so much in the optimizer.

Your Next Step

Build the eval before you touch an optimizer. Pull examples representative of how your system actually gets used, then define what a good output looks like. The scoring criteria is the hard part, it takes longer than building the runner, and it matters more.

Once the eval exists, every other decision, which model, which optimizer, whether a prompt change helped or hurt, has a foundation to stand on.

Not enough time to do it yourself? We do this as a service. Schedule a call with our team today​ to talk about optimizing your model.​

Around the Data Science World

Learn New Data Skills At the ODSA, we don’t believe in generic training or one-size-fits-all degrees. We were built by practicing data scientists who needed more data scientists—so we train people the way we wish we’d been trained. That means real-world skills, taught by professionals who use them every day, in a format designed for working adults... Join us for our Fall Semester: ​Business Intelligence Certificate​ ​Machine Learning Certificate​ ​Data Visualization Certificate​ ​Fundamentals of AI​ ​Agentic AI Engineer Certificate​ APPLY TODAY ​Scholarships are available!

​

We want to hear from YOU

So we may have tried to be too fancy with our feedback form. Turns out, it was skipping all the important questions like "What did you ACTUALLY think of the events?" and "What do you want to see at Omaha DS/AI Week in 2027?"

So, whether you have or haven't filled out the form yet, now's a good chance to share your feedback!

​Take this 1-3 minute survey to help create Omaha Data Science and AI Week 2027!​

If you were there, we want to know how your experience was. If you weren't, we want to know what would bring you next year.

To evaluations and beyond!

NATE WATSON CEO, Contemporary Analysis

Was this email forwarded to you? Join our weekly HIGHLY TECHNICAL Newsletter for your weekly dose of AI, Tech, and Data Science. SIGN ME UP!

WAYS TO WORK TOGETHER

Let us build your Data Solution We help businesses of all sizes to increase visibility, drive more traffic and improve commercial success. ​Schedule a Call ​

Enroll for Classes at ODSA Come learn from 6-9pm, what a data scientist did from 8-5pm” ​Talk to an Advisor about your next steps!

You're receiving this email because you signed up to receive our Data Science resources from Contemporary Analysis or the Omaha Data Science Academy. If you would like to stop getting our weekly newsletter, but stay on our other lists, update your Preferences or Unsubscribe

1402 Jones Street, Omaha, NE 68102