Data Science Salon Podcast

Ben Dubow studied the Middle East during his undergrad and took a job tracking terrorist groups.  After a brief stint at a large tech company, he launched Omelas, a company that combines AI and subject matter expertise to deliver intelligence to national security professionals.In today's episode, our Senior Content Advisor Q McCallum caught up with Ben to learn more about what Omelas is up to and how the company applies AI and data analysis to its mission.Along the way they explore the value of data in context; why it's important to ask the right questions of the right data, and not just the whole pool; the power of involving humans in the data pipeline; and what it takes to do NLP and NER at scale.  The two also talk about the impact of generative AI on democracy and authoritarianism.  A topic which, interestingly enough, holds lessons for corporations that plan to release AI chatbots.Links mentioned in this episode:

• Ben's LinkedIn profile
• Omelas website
• Ben's writing on the Center for European Policy Analysis (CEPA) website
• Article in Les Echos describing the project "Le Monde in English": "« Le Monde » parie sur l'étranger pour stimuler sa croissance"
• Q's write-up on "Risk Management for Generative AI Bots" is available on both his O'Reilly Radar page and his blog.

If you've been in the data game long enough, you've probably seen this before: a stakeholder or product owner approaches you with a project that's 95% done, and they'd like you to … "sprinkle some AI on it."  They've heard that this "AI" thing can be useful so they want some of it in their latest effort.Data scientist-turned-product person Noelle Saldana has experienced the "sprinkle some AI on it" request more times than she'd care to remember.  Our Senior Content Advisor Q McCallum met up with Noelle to explore this phenomenon.  How does this happen? (Hint: "corporate FOMO.")  What should you do when stakeholders insist on implementing AI that isn't actually going to help?  What about when your data scientist peers seem like they're doing this for the sake of "résumé-driven development?"Ultimately, the pair work through the bigger issue: how do you make peace with companies throwing money at AI like this? And how can these companies use this approach to their advantage?As a bonus, Noelle shares how she made the move from a data scientist role into product management.  If this path sounds interesting to you, take a listen.

• Noelle's Data Council talk, "Hot Takes and Tragic Mistakes: How (not) to Integrate Data People in Your App Dev Team Workflows"
• Find Noelle on LinkedIn: https://www.linkedin.com/in/noellesio/
• Q's blog post (which came out much better thanks to Noelle's help): "AI isn't something you just add to a company"

Sometimes the most valuable data IN your company ... is the data LEAVING your company.That's Solomon Kahn's view on data products, as well as the premise behind his latest venture: Delivery Layer.For this episode, our Senior Content Advisor Q McCallum reached out to Solomon to check in on the new startup, and to tap his expertise in the world of data products.Solomon's been at this a while.  He's run high-revenue data products in some notable places, including Nielsen.  Over the years he's learned a lot and we're excited for him to share some of that hard-earned knowledge here on the show.In this extended conversation, the two explore: the reasons why building a data product is different (and, in many ways, more difficult) than building traditional software products; how the people involved can impact the outcome; why a good sense of risk management can make all the difference; and what purple cars have to do with all of this. (No, seriously.  Purple cars.)Along the way, the pair talk about the early days of the data field, and how much it has changed.

• Solomon is active on LinkedIn.  You can follow him for his daily updates at https://www.linkedin.com/in/solomonkahn
• Delivery Layer: https://www.deliverylayer.com/

In this episode, James and Q explore:

• The ideas of risk and uncertainty.
• What is "probabilistic thinking" and why is it important for data scientists?
• The career progression of a data analyst, and what it means to develop statistical acumen.
• Thinking in terms of distributions, and thinking in different moments of a distribution.
• Seeing BI, AI, and simulation in terms of punctuation.  (No, seriously.)
• How to bridge the gap into thinking probabilistically

And, just a reminder: James only speaks for himself in this episode and he does not represent his employer.Links mentioned during our discussion:

• You can find James on Twitter at @cmastication and on LinkedIn at https://www.linkedin.com/in/jamesdlong/
• R Cookbook, 2nd Edition (which James co-authored)
• RenRe's open roles: https://bit.ly/renrejobs
• Q's blog posts on "punctuation in data":  Periods and Question Marks (BI and AI), and  then Ellipses (simulation)

The list of books James mentioned:

• Thinking in Bets (Annie Duke)
• Fortune's Formula (Poundstone)
• The Lady Tasting Tea (Salsburg)
• Fooled by Randomness (Taleb)

We've all heard the term "economist," sure. But exactly what does and economist do? And as economics is a very data-driven field, where does their work intersect with data science, machine learning, and AI?

To answer that question, Senior Content Advisor Q McCallum spoke with Amar Natt, PhD. She's an economist at Econ One Research, and her work focuses on advanced analytics and predictive modeling. Does that sound like ML to you? Well, Amar explains that it's similar in some ways, different in others. From there, she tells us about techniques economists can learn from data scientists, and what data scientists can pick up from econ. (Hint: "causal inference." You heard it here first.) You can find Amar online:

• LinkedIn: https://www.linkedin.com/in/amarita-natt-ph-d-79028313/
• Econ One Research: https://www.econone.com/staff-member/amarita-natt/

Be part of the conversation and connect with the data science community at DSS Miami Hybrid on September 21, 2022.

Book your ticket now.

When people think about The Home Depot, they probably think more about lumber
and tile than they do ML models.  Sure, there is plenty of lumber.  But machine learning also plays a key role in the business, in places that customers can see as well as the behind-the-scenes operations.Senior Content Advisor Q McCallum met up with Pat Woowong, Director of Data Science at The Home Depot, to explore how the company mixes their very rich dataset with domain knowledge to employ machine learning deep inside the business.  To frame this, he walked me through the Falloff model and Lead scoring, two projects that his team deployed to address the unique challenges of a company that handles both retail and services.During our conversation, we discussed: understanding where models fit into the bigger business picture; using expert domain knowledge to drive feature selection and feature engineering; the value of process; and, to top it off, what it's like to work at The Home Depot.Other places to find Pat:

• LinkedIn: https://www.linkedin.com/in/patwoowong/
• "How THD keeps shelves stocked using ML" (the talk he mentioned during our interview): https://twimlai.com/podcast/twimlai/how-ml-keeps-shelves-stocked-home-depot-pat-woowong/
• "The Value Proposition for Using ML in Brick-and-Mortar Retail Stores: Home Depot" https://www.youtube.com/watch?v=rF8jtdX-hGo

Be part of the conversation and connect with the data science community at DSS Miami Hybrid on September 21, 2022.

Book your ticket now.

A lot of data scientists work in the private sector: finance, adtech, retail, and all that.  Today's guest offers her perspective on what it means to do data work in the federal space.In this conversation, our Senior Content Advisor Q McCallum spoke with Dr. Pragyansmita Nayak, Chief Data Scientist at Hitachi Vantara Federal.  They explored how different federal agencies use data and how they share datasets with each other. They also talked about how to measure operational efficiency, when you can't rely on metrics like "profit." And, the big question: should we release t-shirts that read "just give me my AI solution!" ?You can find Pragyan online:

• Twitter: https://twitter.com/SorishaPragyan
• LinkedIn: http://linkedin.com/in/pragyansmita

The book Q mentioned is Army of None, by Paul Scharre.

Charles Irizarry, CEO & Co-Founder at Strata.ai had the chance to talk to Nirmal Budhathoki, Senior Data Scientist at VMware Carbon Black and Moody Hadi, Group Manager - New Product Development & Financial Engineering at S&P Global. Tune in to hear about ML techniques they are using in their current roles, tools to put ML into production, model explainability, and future trends.

In the previous episode, our Senior Content Advisor Q McCallum met with product manager Chris Butler to explore the role of uncertainty and how it relates to AI product management.  That conversation sets the stage for Chris and Q to talk about communal computing today.

Chris starts by explaining what shared, AI-backed devices mean for data collection, analysis, and regulation. After that, Chris and Q explore important questions such as: What are some challenges in getting communal computing devices to coordinate?  How do social norms mix with assumptions made by the ML models behind these devices?  What do we lose when we use data lakes? How do product managers and machine learning engineers interact on these kinds of projects? What do communal computing devices have in common with software developers on shared platforms?And, most importantly: what does all of this have to do with the film Napoleon Dynamite ...?

• Chris has published a series of articles on communal computing: Communal Computing intro, Communal Computing’s Many Problems, and A Way Forward with Communal Computing.
• You can also watch some of Chris’s communal computing talks:
• AIxDesign Communal Computing workshop with animistic design mapping
• Bots and AI Meetup - Communal Computing - Solving multi-user Alexa and Google Assistant use cases

Formulated.by’s Senior Content Advisor, Q McCallum, caught up with Linda Liu (Hyrecar) and Giacomo Vianello (Cape Analytics).  Our guests explored the techniques and tools for the various data projects they are running, some of the challenges of working with geospatial data, and how their companies approach data-related research efforts.

This discussion also explores the context around which we collect data, polysocial reality, design individualism, and contextual integrity.  (Yes, we covered a lot of ground in just 45 minutes.)

Because of our tight schedule, Chris and Q had to stop before they could get to their second topic.  That’s why Chris will be back in the next episode to talk about communal computing and what that means for AI.            

bio: Chris Butler is a product manager, writer, and speaker with over 20 years of product management leadership at Microsoft, Waze, KAYAK, and Facebook Reality Labs. He facilitates critical decision making for teams that build new and innovative products and created techniques like Empathy Mapping for the Machine and Confusion Mapping to create cross-team alignment while building AI products. 

He is now Assistant Vice President, Head of Product Operations, at Cognizant where he PM’s the PM experience.Learn more about Chris and his work through his Linkedin, Twitter, by reading some of his articles on Medium or watching some of his past talks on YouTube.

• Article: “Our Favorite Questions”
• Chris’s Product Mindset talk from Cognizant Softvision Product Day
• Chris’s Prototyping for AI/ML at the AI for PMs Summit
• At the end, we mentioned Chris’s articles on communal computing.  You can read these to get a head start on our next episode:
• Communal Computing intro
• Communal Computing’s Many Problems
• A Way Forward with Communal Computing
• AIxDesign Communal Computing workshop with animistic design mapping

Formulated.by’s Senior Content Advisor, Q McCallum, caught up with Vidhi Chugh (Walmart), Piyanka Jain (Aryng), and Tempest van Schaik (Microsoft).  Our guests explored the impact of the Covid-19 pandemic on hiring and retention, then shifted to a discussion on finding and serving as a mentor.

Formulatedby's Senior Content Advisor, Q McCallum, met up with Sonali Syngal to explore these questions.  Sonali is currently a data scientist at MasterCard and is about to join the team at Expedia.  She came to data science from the rather uncommon entry point of economics. In this episode we see that her career path has given her key insights on how to join this field and what are the differences between the various roles therein.(Some listeners may recognize Sonali's voice from a previous episode: she spoke at Data Science Salon in December 2020, where she also joined us for our Coffee Chat.  You can check out that episode to learn more about Sonali's take on ML/AI in the world of fintech.)

• Sonali mentioned Christopher Olah's blog, which lives at: https://colah.github.io  (Christopher was working at Google when Sonali found his blog, though he has since moved to OpenAI.)
• Sonali and Q are both fans of the 3Blue1Brown YouTube channel.  It offers clear, relatable explanations of linear algebra and neural networks: https://youtu.be/aircAruvnKk
• This is DeepLizard’s deep learning playlist that Sonali mentioned: https://youtube.com/playlist?list=PLZbbT5o_s2xq7LwI2y8_QtvuXZedL6tQU

What was it like to move from a physics lab into the data scientist's chair?  How did she find that first job? And what elements of her PhD experience have proven especially valuable in her machine learning work?  Join us in this conversation to find out.

• Dr Sarkar works on machine learning at Slice . (She's hiring!)
• You can also find her on LinkedIn.

Micheline has more than twenty years' experience at the intersection of data and money, and has been a Chief Data Officer (CDO) with 3 different organizations, leading and scaling data strategy, infrastructure, and platforms. She also led data commercialization efforts at Ford.  Her career includes early data brokers, automotive and logistics companies, financial services and insurance, health care, and energy.  Oh, and then there was that stint as the CDO of the Federal Reserve.  She's a real powerhouse in the data field and we're very happy that she was able to join us.

• You can find Micheline on LinkedIn and Twitter
• The paper Q mentioned early in this episode is Business Models for the Data Economy (McCallum, Gleason)
• The book Micheline mentioned near the end is Doug Laney's Infonomics: How to Monetize, Manage, and Measure Information as an Asset for Competitive Advantage

Data scientists and ML engineers write a lot of code: building data pipelines, wiring up models, and sometimes translating concepts from research papers into algorithms.  

Once in a while, that code runs into performance problems.  These can be painful to debug when you don't come from a formal software development background.  That's why Formulatedby's Senior Content Advisor Q McCallum rang up Matt Godbolt to learn the deep details of software testing, tracing performance bugs, working with data at scale, and how data scientists can work with developers to prepare their code for a production handoff.

Matt Godbolt has more than 30 years' experience writing code.  He's spent most of that time working in the performance-focused environments of console video games, high-frequency trading (HFT), and algorithmic trading.  Matt is the creator of the Compiler Explorer website, and also co-host of the Two's Complement podcast.

(Note from Q: My audio is a little choppy, but Matt's is perfect.  And you're here to hear him, anyway...)

Matt and Q mentioned a few links during their talk:

• Michael Abrash’s Zen of Code Optimization
• Brendan Gregg’s Flame Graphs
• Matt's blog and videos
• Related to the discussion on performance enhancements in Quake, Matt has a video on “how Wolfenstein worked”
• Compiler Explorer