Mar 20, 2014 · 26m · mad
Jesse St. Charles, Knewton // Data Driven #25 // March 2014 (Hosted by FirstMark Capital)
Jesse St. Charles · 20m spoken Andy Beveridge · 57s spoken Claire Wyckoff · 18s spoken Matt Turck · 16s spoken Kevin Noonan · 7s spoken
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this Data Driven presentation, Knewton's lead data scientist Jesse St. Charles explains how adaptive learning platforms utilize knowledge graphs, real-time predictive modeling, and cross-functional team structures to personalize education at scale. He details Knewton's B2B partner ecosystem and highlights empirical data demonstrating improved learning outcomes across diverse academic subjects.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Matt holds 1.2% of the talking time here. How this is scored →
Matt as informed peer 0.2 Guest teaching 2.2 Guest disagreement 0.7 Matt pushing back 0.2
speaking balance: gold is Matt, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 20:06 Audience challenge on cross-domain platforms
Audience member Andy Beveridge forcefully challenges Jesse's core premise, asserting that a single platform cannot effectively generalize across fundamentally distinct domains like history and math.
Hardest push from Matt ▶ 24:42 Host playful interjection on monetizationHost Matt Turck playfully pushes back on a business model question by interjecting 'It's not about money' during audience Q&A.
Biggest teaching moment ▶ 21:20 Reframing history as prerequisite knowledge graphsJesse educates the room on how subjects typically viewed as rote memorization, like history, actually map into structured networks of prerequisite event relationships.
Matt holds his own ▶ 0:00 Host opening framing of KnewtonHost Matt Turck demonstrates domain awareness by introducing Jesse as a leader in massively adaptive learning platform technology.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Matt as informed peer | Guest teaching | Guest disagreement | Matt pushing back | Why |
|---|---|---|---|---|---|---|
| Addressing Problems in the Industrial Education Model | 0 | 2 | 1 | 0 | Jesse introduces the core problems in the industrial educational model and Knewton's adaptive platform approach. Host scores are zero because this is a solo presentation monologue with no host interaction. | |
| Data Foundations and Core Entities at Knewton | 0 | 2 | 0 | 0 | Jesse details Knewton's data foundation, including learners, contextualized textbooks, and knowledge graphs. This is a monologue presentation segment, resulting in zero scores for host activity. | |
| Visualizing Student Trajectories Through Knowledge Graphs | 0 | 2 | 0 | 0 | Jesse presents video visualizations illustrating student learning paths and real-time ability estimations on knowledge graphs. Host scores remain zero as it is entirely a guest presentation. | |
| Inferential Products and Measured Educational Outcomes | 0 | 2 | 0 | 0 | Jesse discusses Knewton's inferential products, validation hurdles with human behavioral data, and internal efficacy studies. Host scores are zero due to the monologue structure. | |
| Data Empathy and Cross-Functional Inferential Team Structure | 0 | 2 | 0 | 0 | Jesse explains hiring criteria for data scientists, introducing the concept of data empathy and cross-functional team structures. Host expertise and pushback are zero in this solo talk. | |
| Presentation Conclusion and Transition to Q&A | 1 | 3 | 3 | 1 | Jesse concludes his talk and host Matt Turck opens the floor for audience Q&A, where professor Andy Beveridge challenges Knewton's cross-domain platform assumption. Jesse defends the platform model by explaining how non-STEM subjects like history rely on prerequisite knowledge networks. |
Statements from this episode (15)
St. Charles: Industrial education's one-size-fits-all model leaves everyone underserved
“It is one size fits all. And to me that means that everyone's underserved.”
St. Charles: Student interaction data in educational software goes completely unused
“Students are already generating all this data as they're interacting with these educational products, but they're just falling on the floor. We're not doing anything with it.”
St. Charles: Knewton operates as an API for content partners
“We're actually an API, and we, we're using that API to provide the above capabilities to any content partner that decides to work with us.”
Jesse St. Charles: Knewton served roughly 4 million students in 18 months
“We've had roughly four million students on the platform in the past year and a half.”
Jesse St. Charles: Knewton has systemized over 200 textbooks
“We have over 200 textbooks that have been described to our system.”
Educational knowledge graphs encode expert prior beliefs about knowledge structure
“So from a data science perspective, one way to think about this, and it's the way I think about it is that this graph is actually encoding expert prior belief about the structure of knowledge.”
Knewton uses hierarchical agglomerative clustering to refine knowledge graphs
“Here we, we've done some hierarchical agglomerative clustering to detect the relationships between pieces of content using performance data.”
Knewton saw nearly half a grade improvement with a major partner
“But at one of our largest partners, we've seen nearly half a great improvement as usage of adaptive assignments increases.”
St. Charles: Student proficiency is straightforward to validate in edtech
“Estimating student proficiency, relatively straightforward. We can validate that because we actually see that the model itself predicts outcomes.”
St. Charles: Knewton needs partner investment to run controlled studies
“And also because we're a platform, we have to coordinate with partners to actually impact users. And so this gets back to what I was saying before about some of the challenges of running control studies. We actually require explicit and investment by our partn…”
St. Charles prioritizes 'data empathy' over domain expertise when hiring
“Something that I personally really value in the people that I hire is something I call data empathy, and it's a skill.”
Beveridge: Edtech platforms cannot easily span fundamentally different academic domains
“I don't really believe that You can go across these domains with a platform necessarily”
St. Charles: Modeling conceptual dependencies is beneficial across all subjects
“There is requisite knowledge that you need, and that's where we're trying to model the content that we're working with, and so I guess it's, what, the question is whether or not you buy that there exists requisite relationships in content, and whether respecti…”
St. Charles: Knewton is deployed in both higher education and K-12
“We're deployed in Higher education institutions power a lot of Pearson publisher books which are, and a lot of higher ed. Also we actually just launched a product with Houghton Mifflin Harcourt, which has K-twelve. So we're actually in both now.”
St. Charles: Knewton monetizes by taking cuts of textbook sales and courses
“We're basically providing a value add for every time a piece of that content gets distributed, and so so we might take a cut of the textbook sale It's a digital textbook or real textbook sale if there's a digital component to it or a cut of the actual of a cla…”