Podcast Episode Summary
Harvard Data Science Review Podcast
Episode: Tracking the Most Intoxicating Data: A Conversation With Eric LeVine
Date: November 20, 2025
Guest: Eric LeVine, Founder of CellarTracker
Hosts: Liberty Capito (A) & Shali Meng (B)
Episode Overview
This episode delves into the unique intersection of fine wine and data science through the lens of CellarTracker—the world’s largest wine tracking and discovery platform. Eric LeVine, CellarTracker’s founder, shares how his personal passion evolved into a community-driven, data-rich platform, and discusses how user-generated data, AI, and machine learning are transforming both wine appreciation and the wider field of recommendation systems. The conversation covers the challenges and opportunities of translating subjective human preferences into actionable recommendations, as well as the importance of privacy, user experience, and data quality.
Key Discussion Points & Insights
1. The Accidental Genesis of CellarTracker [02:07]
- Eric LeVine traces the roots of CellarTracker to a combination of his work at Microsoft on crowdsourced error reporting and a transformative wine experience in Tuscany.
- “CellarTracker was an accident that started in 1999… My day job is dealing with mining and basically crowdsourced big data for software reliability, and I’m keeping this growing cellar in a spreadsheet … It's a great tool. It's a crappy database.” [02:07]
- What began as a personal tool quickly spread to friends and eventually evolved into a global community with over a million users and 200 million bottles tracked.
2. How Users Actually Engage with Wine Data [05:19]
- The platform’s primary value is as a productivity tool for cataloging collections, but its broader utility is as a research and discovery engine for millions more.
- “About 8 million a year is not to catalog their cellar, but to go and search and research all the reviews.” [05:19]
- Data reflects staggering value—users have cataloged billions of dollars' worth of wine over time.
- There's wide diversity in user opinions and preferences, and surprisingly high engagement for some power users.
3. The Subjectivity of Taste and the Promise of AI [09:01]
- The subjectivity of wine, Eric says, creates opportunity for AI and data science to test their limits.
- “Wine is completely subjective ... but that actually creates an enormous opportunity for us to test all these AI tools.” — Shali Meng [09:01]
- CellarTracker experiments with recommendation models—seeking “digital twins” and summarizing community reviews to answer, “Will I like this wine?”
- Challenges in getting AI to distinguish strong user dislikes: “It is really difficult to get these models to not say something nice.” [11:53]
- A memorable test case: Most users wanted to test whether the AI thought they’d like “Caymus,” a polarizing California wine.
4. Balancing Human Element and Machine Learning [15:31]
- Eric discusses the importance of centering recommendations around exploration, not just commerce-driven popularity.
- “If people are looking for hey, show me something new or help me find that hidden gem, then that's the trick.” [15:31]
- The vast diversity and “long tail” of wine data presents both a challenge and an opportunity.
5. Data Quality, Collection, and Cleaning [18:54]
- The platform faces three dimensions of data quality:
- User Tracking – Users may not update their cellar accurately, so efforts are in place to lower friction and automate tracking (e.g., computer vision for label recognition).
- Database Integrity – With 5 million wines, new entries are curated and merged to reduce duplication and errors.
- Preventing Manipulation – Rare issues with fake reviews are quickly identified and reported by the community.
- “Generally it’s in the rare cases when people have posted fake reviews … it sort of stands out like a sore thumb.” [18:54]
6. The Language of Wine & NLP [22:48]
- Wine’s niche vocabulary (“bruised pear,” “forest floor”) creates barriers for novices and challenges for NLP.
- “Wine's greatest strength, its incredible diversity ... also creates such a barrier to entry. Until you've really dove in and tasted some wines or had sort of a seminal experience, it's just a scary topic for people.” [23:25]
- CellarTracker strives to help users gain confidence in their own palates, regardless of their ability to use descriptive language.
7. Privacy and Ethical Use of Data [25:39]
- Eric underscores a strong “no creepy” policy for user data, allowing for anonymous use and collection privacy, and prioritizing user control.
- “The phrase I use to the team is no creepy. … let’s be just above board.” [25:39]
- Public notes foster community but private options exist.
8. Innovations in Wine Discovery and Commerce [28:37]
- CellarTracker aims to be a research/discovery tool not just at home, but at retail and restaurants.
- Working on features like “Restaurant Hero,” which recommends wines from a restaurant’s list based on user taste and value.
- Envisions disruption of the current email-heavy wine sales model through personalized, privacy-centric offers.
- “I would love a world where we are ingesting all of that [wine offers] and running it with the user's control on our platform ... Coming back with, here’s the hot list of what’s being offered right now that’s actually relevant to you.” [30:28]
9. The Future of AI and Wine [32:20]
- AI will further democratize wine appreciation—making research, exploration, and commerce easier, and potentially helping producers adapt to climactic challenges.
- “AI is going to change the world in this space for the good period and a lot of dimensions.” [32:20]
Notable Quotes & Memorable Moments
-
On technological serendipity:
“My mind was blown. And in tech speak, I kind of had my bit flip moment.” – Eric LeVine [02:07] -
On the scale of wine data:
“Every year, a billion to a billion and a half dollars of wine ... 25-ish billion in total over time. So it's just like, wow.” – Eric LeVine [05:19] -
On AI’s limitations in taste prediction:
“It is really difficult to get these models to not say something nice.” – Eric LeVine [11:53]
“When we started rolling out this tool in beta, like, 90% of the people who were trying to use it... Everyone put in Caymus because it's such...” [11:54] -
On user engagement:
“There are users ... spending more than a thousand hours a year aggregated on the site.” – Eric LeVine [05:19] -
On wine’s intimidation factor:
“I think everyone, I just want them to have the confidence to answer the most basic question: did I enjoy this?” – Eric LeVine [23:25] -
On ‘no creepy’ data:
“Let’s be just above board ... don’t do anything creepy with users' data.” – Eric LeVine [25:39] -
On his personal wine ‘magic wand’:
“Lapand ... I can actually still taste and smell it now. And that was 2010, so 15 years ago. So sorry. All respect to the Petrus guys, it's La Pan.” – Eric LeVine [33:56] -
On his ‘last wine’:
“...the area that I’ve really appreciated the most over the last probably 10 years is the southern Rhone Valley. In particular, Chateauneuf du Pape ... Domaine de Pegau.” – Eric LeVine [35:23]
(with a joyful fist pump)
Key Timestamps
- [02:07] — Eric’s journey: Microsoft, Tuscany, and the birth of CellarTracker
- [05:19] — Platform’s dual use: cataloging vs. research; surprising statistics
- [09:01] — On subjectivity, data science, and the quest for predictive taste
- [11:53] — The AI challenge: recommendations, “digital twin”, and Caymus test case
- [15:31] — Recommendations, avoiding commerce bias, tackling the ‘long tail’
- [18:54] — Data quality: user error, database curation, preventing manipulation
- [22:48] — NLP and wine descriptors: helping novices break through the intimidation factor
- [25:39] — Privacy-first principles and “no creepy” data use
- [28:37] — Future tools: restaurant recommendations, commerce disruption
- [32:20] — The potential of AI to reshape wine appreciation and the industry
- [33:56] — “Magic wand” question: if Eric could have any wine
- [35:23] — “Last wine” question and Eric’s passion for Chateauneuf du Pape
Tone & Takeaway
The conversation is characterized by Eric’s humility, passion, and commitment to user experience and community. He emphasizes the joy of discovery and inclusivity in wine appreciation, while candidly sharing the technical, cultural, and business complexities involved. The episode invites both wine novices and data enthusiasts to appreciate how personal taste is becoming part of the new frontier of personalized, responsible, and user-empowering technology.