Mar 15, 2021 · 29m · mad
Fireside Chat: Jack Hanlon (VP Data, Reddit) with Matt Turck (Partner, FirstMark)
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 fireside chat hosted by FirstMark's Matt Turck, Jack Hanlon, VP of Data at Reddit, shares insights on scaling Reddit's data organization, engineering robust hybrid infrastructure, and advancing ethical AI and privacy practices.
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 14.9% of the talking time here. How this is scored →
Matt as informed peer 3.1 Guest teaching 4.6 Guest disagreement 0.4 Matt pushing back 0.3
speaking balance: gold is Matt, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 23:01 Blunt dismissal of AWS Redshift architecture
Guest forcefully rejects AWS Redshift as inadequate for Reddit's needs, plainly stating that its core architecture has failed to keep up with modern data demands.
Hardest push from Matt ▶ 7:16 Host presses guest on company headcount contextHost interrupts to explicitly question and clarify the scale of Reddit's total company headcount relative to its single data engineer at the time.
Biggest teaching moment ▶ 24:35 Masterclass on detecting and mitigating algorithmic data biasGuest provides an in-depth breakdown on systemic data collection bias, referencing Carolyn Criado Perez's book Invisible Women and outlining practical auditing frameworks.
Matt holds his own ▶ 13:35 Host's domain expertise acknowledged by guestHost demonstrates deep knowledge of data lineage and market trends, prompting the guest to explicitly defer to Matt's superior market knowledge on venture startup behavior.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Matt as informed peer | Guest teaching | Guest disagreement | Matt pushing back | Why |
|---|---|---|---|---|---|---|
| Jack Hanlon's Career Path to Reddit | 1 | 2 | 0 | 0 | Host opens with a standard, open-ended question about the guest's career progression to Reddit. Guest shares a non-linear path from music major through sales, ad tech startups, and Jet.com to advising Reddit. | |
| Overview of Reddit's Four Data Organizations | 2 | 4 | 0 | 0 | Host asks organizational questions about how data is structured at Reddit. Guest breaks down the four core data groups and the hub-and-spoke matrix model. | |
| Roadmap Prioritization and Scaling Data Engineering | 3 | 5 | 0 | 1 | Host probes into prioritization and interrupts to clarify headcount numbers. Guest reveals the surprising fact that Reddit had only one data engineer managing billions of daily events when he joined. | |
| Reddit's Data Engineering Stack and Quality Tools | 5 | 5 | 0 | 1 | Host asks targeted questions about Reddit's tech stack across BI, notebooks, and data lineage. Guest acknowledges the host's market expertise regarding venture-backed data startup pressures. | |
| Great Expectations Unit Testing and Streaming Pipelines | 4 | 4 | 1 | 0 | Host relays technical audience questions regarding Great Expectations and ETL pipelines. Guest playfully notes the audience picked his weakest area before explaining data unit testing principles. | |
| Machine Learning Applications, Personalization, and Anonymity | 4 | 6 | 0 | 0 | Host prompts guest on specific machine learning applications and infrastructure choices like Imply and Druid. Guest explains how Reddit's unique conversational corpus was leveraged by major AI labs while preserving user anonymity. | |
| Infrastructure Migration from AWS Redshift to GCP | 4 | 7 | 2 | 0 | Host relays audience questions on cloud migrations and AI bias. Guest gives a candid evaluation rejecting Redshift's capabilities and offers an insightful lecture on framework strategies for preventing machine learning bias. | |
| Future Data Trends, Ethical Design, and Conclusion | 2 | 4 | 0 | 0 | Host transitions to quick closing questions on industry trends and recommendations. Guest compares data product engineering ethics to civil engineering standards for bridge safety. |
Statements from this episode (12)
Hanlon: Reddit uses a hub-and-spoke model for data science
“Data science is hub and spoke. So so effectively that is all of those folks report into the director of data science, who's, who's one of my team members and, but they actually, when we're in a, when we're in an office setting, they sit with the downstream tea…”
Hanlon: Reddit grew from around 600 to 800 employees in 14 months
“When I got to Reddit overall, I'd say Reddit was maybe 600 people 14 months ago. Right now, oh, maybe 800.”
Hanlon: Reddit's data team will probably reach ~120 people by end of 2021
“By the end of the year, the data organization I just described will be probably the third largest organization at Reddit with a 120 or so people.”
Hanlon: Reddit ingested 55-60B daily events managed by one person in 2019
“I came in, we were doing probably about 55 or sixty billion events a day into a data warehouse that one person was managing.”
Reddit runs production on AWS and analytics/ML on GCP
“From the top view, I'd say we're AWS prod with originally a lot of Postgres For a variety of prod systems, we actually transit almost all that stuff over to GCP, and we use TensorFlow and BigQuery for our analytics and our model concerns”
Data startups push horizontal platforms, but enterprise buyers want single features
“The companies I see, it feels like a number of the startups I see in the space get pressure to expand to be more horizontal solutions. And then frankly, like we want them for one thing, but not for the three other things that they want to do.”
Reddit isolates malformed data events in an automated penalty box
“You can do stuff for event validation to say, ah, events that don't match certain things get dropped and sort of get put in the penalty box, which we've done and said, okay, developers, you can't actually join these to anything. You can't actually use them if …”
Reddit is leaning heavily into becoming a Kafka shop for streaming
“So, you know, I think we're ending up leaning heavily into being a Kafka shop, building more of the stuff into base plate, into more of these core services”
Google, Facebook, Microsoft, and OpenAI used Reddit data for AI models
“Google, Facebook, Microsoft, and OpenAI all used Reddit's data to train their conversational AI models.”
Jack Hanlon: 2020 was the first year every Reddit surface implemented personalization
“Last year was the first year that every Reddit surface saw personalization.”