» teej @teej_m Twitter profile

Pinned Tweet

» teej

@teej_m

2 years

I thought these were drawn exclusively for O’Reilly. My whole life is a lie.

376

4K

36K

Last Seen Profiles

@michael_dove16

@JohnstoneYT

@TreoloMichael

@KEIENDO_jp

@Jel__official

@caro_milanesi

@Iam_OmobolanleE

@Tejumola__

@Iris_Descent_

@russbites

@jangkyuyin

@TyeHulud

@LokiAndGambit

@NCLU_ORG

@PeterraynorSFW

@BHSAthletics

@eva_lequeen

@StangeMaria

@clynnnnn5

@NM_NANAPARTY

@semiyann

@Serapen1419795

@gis_fluor

@SophiaMoore2025

@Kyrokana

@TheRealMizVic

@iamJDCrafton

@NaturePhysics

@itsnotbobitsbAb

@KeananLocke

@1DBRACOM

@LionOfUldah

@Realajbarnes

@UNO650AM

@MicoJo2

@janfrodeno

» teej

@teej_m

10 months

There’s only one 𝕏 that I respect

20

1K

5K

» teej

@teej_m

3 years

“We have big data, can you help us?” The data:

44

588

4K

» teej

@teej_m

4 months

@chrisalbon Every single software company CFO saw what happened with Twitter (90% labor cost reduction) and started asking hard questions of leadership. Also section 174.

18

16

995

» teej

@teej_m

10 months

@anandc Can we not glorify CEOs committing crimes

6

15

946

» teej

@teej_m

4 years

Building a data pipeline in 2020 is like building a bridge in the 14th century • You do a lot of work that gets thrown away • Half the job is getting rid of the stuff you dont want • The folks who started it are dead by the time it's done

12

153

924

» teej

@teej_m

1 year

I have done hundreds of analytical SQL coding interviews. My standards are higher than most. Here’s what you need to know to pass it.

b bennett | 500+ connections

@thebmbennett

1 year

i’m coaching someone through a data analyst job search, and when i asked if they knew sql their response was “i dont need to learn it bc it’s easy to pick up on the job” sql is deceptively complex. easy to pick up. difficult to master. please take the time to learn sql

15

17

335

14

99

856

» teej

@teej_m

2 years

I can confirm @chrisalbon ’s book cover animal is from the 1885 tome Animate Creation by Rev. J. G. Wood I do not have matching bottle of wine

3

9

518

» teej

@teej_m

9 months

I have a teeny little announcement: I've worked with cloud data warehouses for 9 years. And for 9 years, I've felt frustrated at the dev experience. To fix it, I'm working on a new open source project. I'm calling it Titan. Let me give you a sneak peak at what Titan is – 1/

19

52

476

» teej

@teej_m

2 years

This is the story of how I independently landed on the same ideas that make up dbt today. I joined a DTC startup in 2013. This is how their data infrastructure looked shortly before I inherited.

15

53

437

» teej

@teej_m

3 years

Snowflake released MATCH_RECOGNIZE. I'd never heard of it before. Say you run an ecomm store and want to analyze the shopping funnel. To model the steps in SQL you either need a messy set of JOINs or a daisy-chain of window functions. Or you write 15 lines of MATCH_RECOGNIZE:

21

63

425

» teej

@teej_m

2 months

Hey friends, I have news to share! I started a company, it's called Titan Systems. Titan builds security software for Snowflake, starting with access management: users, roles, and permissions. I joined the Y Combinator Winter 2024 batch to help me bring this idea to market.

36

26

395

» teej

@teej_m

8 months

@MichaelFilipiuk @soychotic It allows Walgreens to sell ads. They get to charge more for a 6ft tall video ad than just for shelf placement alone.

3

1

351

» teej

@teej_m

2 years

Databases like Postgres use indexes. If we query a single row in a table, adding an index makes it fast. Snowflake doesn't have indexes. But there are other features that help speed queries up: 1/ Pruning 2/ Clustering 3/ Query Plan Rewriting Let's look at each.

Erik Bernhardsson

@bernhardsson

2 years

@yonidavidson Isn't Snowflake's approach just "use full table scans for everything"?

4

0

10

7

41

317

» teej

@teej_m

2 years

I don't know who needs to hear this – Negotiating job offers is one of the most important professional skills you can learn. I don't know how to teach it, but I at least want to share what it can look like. I sent this exact email to negotiate a prior offer.

11

18

311

» teej

@teej_m

2 years

Congrats! You decided to roll your own data pipeline from Postgres to a cloud data warehouse! I'm proud of you, I think cloud data warehouses are very cool. Here's every question you'll probably need to answer to build this pipeline –

8

33

293

» teej

@teej_m

2 years

@ellewasamistake

0

9

287

» teej

@teej_m

1 year

Bunch of thoughts on data I want to write down so I can start next year fresh. Let’s go.

7

44

280

» teej

@teej_m

2 years

For analytical SQL writers, leetcode SQL questions are way too easy. The most popular hards are solved with row_number, self join, or count-case-when. Beginner stuff. Give me your hardest SQL interview questions. I want to suffer.

35

19

279

» teej

@teej_m

2 years

How does the Snowflake query optimizer work? It's complicated.

Understanding the Snowflake Query Optimizer

The job of a query optimizer is to reduce the cost of queries without changing what they do. Optimizers cleverly manipulate the underlying data pipelines of a query to eliminate work, pare down...

teej.ghost.io

6

43

271

» teej

@teej_m

4 months

SQL

Christopher Ehrlich

@ccccjjjjeeee

4 months

Which technology is this?

892

71

1K

3

18

261

» teej

@teej_m

2 years

I once applied for Head Of Analytics at a startup. I was dismissed quickly, they wanted a CS PhD to solve the bin packing problem. It turns out — they had zero diligence with discounts, spent all their cash on bad user acq., and drowned in retention problems. Layoffs ensued.

6

25

259

» teej

@teej_m

9 months

@paul_rietschka The pandas api is bad

30

2

241

» teej

@teej_m

2 months

@chrisalbon @cursor_ai Ask and ye shall receive - how I use Cursor's codegen

18

15

241

» teej

@teej_m

2 months

Every “business person” who doesn’t understand data tomorrow: Wow, February sales are up 3.6% year over year!

6

15

240

» teej

@teej_m

3 years

I’m barely 15 min into CMU’s advanced DB course, this is wild • He is lecturing from Amsterdam’s red light district!! • It looks like it’s night but he doesn’t have a hotel • The head TA is called as “vicious AF” and is a reformed gang member

12

17

240

» teej

@teej_m

3 years

@TechEmails “If the plan is clear, no meeting is needed” 😍😍

1

5

229

» teej

@teej_m

5 years

I won’t stop beating this drum - SQL is a critical skill and an important programming language.

8

52

223

» teej

@teej_m

9 months

I didn't have pandas coming for me on my 2023 bingo card but here we are

pandas

@pandas_dev

9 months

We are open source, make it better. ;)

308

2K

27K

10

5

215

» teej

@teej_m

1 year

Professional news - I'm on sabbatical! I plan to unwind, go outside, and think about the impact I want with the next 10 years of my career.

20

0

216

» teej

@teej_m

2 years

If you didn't make it to Snowflake Summit, you missed my talk on funnels! "Funnel analysis" is an old marketing term adopted by web & app analytics to measure cohort behavior, onboarding dropoff, and navigation paths. Measuring funnels in SQL is slow and painful. Let's fix that

5

18

213

» teej

@teej_m

4 months

@ClearHueBlue @chrisalbon When have ✨consequences✨ ever stopped a public company executive from making a bad decision?

1

6

205

» teej

@teej_m

1 year

If you work in analytics, I want you to do this today – Go to your boss, tell them you need to expense $90. Tell them it will save you 10 hours of work in the next month, but $90 pays for the whole year. Here's a few of my top hits from Olga's substack.

Olga Berezovsky

@OlgaBerezovsky

1 year

When you go to a little fun data event and run into founders and friends, NY analysts-bloggers (the big one), and meet the best people. I love SF❤️

2

0

19

3

13

205

» teej

@teej_m

4 months

If you report week-over-week metrics to me on the 1st week of January, you're fired. Right to jail, right away.

12

10

199

» teej

@teej_m

1 year

A physical "timestamp"

7

21

179

» teej

@teej_m

9 months

I am excited to announce the developer alpha of Titan! I have one goal: to simplify how you work with data warehouses, starting with Snowflake. Find it on GitHub:

GitHub - Titan-Systems/titan: Snowflake infrastructure-as-code. Provision environments, automate...

Snowflake infrastructure-as-code. Provision environments, automate deploys, CI/CD. Manage RBAC, users, roles, and data access. Declarative Python Resource API. Change Management tool for the Snowfl...

github.com

9

15

179

» teej

@teej_m

8 months

Ok I don’t know why I’m so fired up about this, but – If you run a data science team and your top priority is “automating” the weekly business review, you are not gonna make it. Very negative signal. Your job is not schlep reduction. Your job is business maximization.

Luke Metro

@luke_metro

8 months

After watching friends at FAANG I must admit that management consultants were right about everything. Designing a promo system and org chart aligned with biz needs may be the most important thing that a company does. Principal-agent problems destroy lives. Becoming McKinseypilled

43

122

2K

16

10

164

» teej

@teej_m

8 months

@nicko_md @powerbottomdad1 Europeans referring to where someone was born is so typically European

6

1

159

» teej

@teej_m

2 years

For the last time Frank, you cannot compare this timestamp field across planets unless you’ve adjusted for time dilation. I know the sales numbers don’t add up! I told you before we need ACCURATE position data. I’ve asked the ops team three times to replace the atomic clocks…

CC🥥

@siliconion

2 years

Imagine dealing with time zones between earth and mars. Glad I'll be dead by then

7

4

82

1

7

163

» teej

@teej_m

4 years

I've done this at 5 different co's now. Here's what I do – - Data team reports the weekly metrics - Split into A) notable trends, B) watching close, and C) business as usual - 95% of metrics fall into C and are not discussed I've earned the authority to say "wait and see"

andrew chen

@andrewchen

4 years

"Why did metric X go down last week?" <-- very common question that you might find yourself asking. Or your ceo/board asks you. It's frustrating to not know. But the reality is that there's often random spikes of usage, seasonality, and other external factors

13

28

326

3

11

161

» teej

@teej_m

2 years

In Vegas, Snowflake announced their big new thing: Snowpark for Python. There's just one problem. No one knows what the heck it actually is. new blog, who dis Let's talk about the 5 new things introduced with Snowpark.

5

32

159

» teej

@teej_m

9 months

Aww, apology accepted. Look, we’re on the same team, at least we’re not writing JavaScript.

pandas

@pandas_dev

9 months

We'd like to apologize to @teej_m . This tweet was never meant as a personal attack, or a "fix it yourself" as many users understood. We welcome all criticism and work with the community every day to make the API better, constrained by being as backward compatible as possible.

31

18

733

4

155

» teej

@teej_m

2 years

Great question from Locally Optimistic Slack: Should I split my team of data scientists into those who work on forward-looking insights and others who handle day-to-day adhoc requests? No. The whole team must do both. Let's talk about why.

5

24

155

» teej

@teej_m

3 months

“The data is clean, don’t worry” The data –

Tyler Glaiel

@TylerGlaiel

3 months

224

3K

21K

7

14

148

» teej

@teej_m

9 months

@pandas_dev I tried to fix the most popular page of Pandas documentation so the example was actually usable and yall closed the PR. Lots of people told me they were confused by this.

» teej

@teej_m

2 years

They closed my PR. Pandas is such an important library. If they can’t take small changes I don’t know how they expect to improve the documentation long term. I won’t waste my time with it.

6

1

52

2

3

141

» teej

@teej_m

2 years

dbt v1.3 will include Python models. You can now write models using dataframe syntax that’s more familiar to those who use pandas. A major objection from @josh_wills - switching from SQL to Python is too much work to be worth it. Well … I may have solved that problem.

Josh Wills

@josh_wills

2 years

@teej_m the expression on the right is gorgeous, but I feel like the amount of boilerplate and rewriting required to get to it from my starting SQL query would preclude me from doing the work in all but the most extreme cases

2

1

7

9

16

141

» teej

@teej_m

2 years

I have committed the cardinal sin of programming. I wrote a package manager. Please forgive me for what I've done.

4

6

141

» teej

@teej_m

4 months

@LIM49Spartan Wouldn't this presumably have shown up in a simple audit of the hypergolics supply? If you ship enough PBVs with water in them, surely you have too much expensive gas left over. The mistake had to be known sooner by somebody.

7

2

141

» teej

@teej_m

2 years

DuckDB is a new embedded database (like SQLite) that's built for analytics. Their goal is no external dependencies. That makes it simple to drop-in to all sorts of places, like a mobile app, a Jupyter notebook, or your smart toaster. That effort is paying off, check this out -

Hamilton Ulmer

@hamiltonulmer

2 years

excited to share an alpha version of @RillData Developer, a reactive EDA-centric SQL tool for exploring and transforming datasets. Something I've always wanted as a data person :) Powered by @duckdb and @sveltejs ✨ try it & let me know what you think!

15

47

331

3

23

141

» teej

@teej_m

7 months

Cool paper out of Stanford/Airbnb – 2-sided marketplaces are a pain to A/B test. It's impossible to stop test cells from interacting. That causes interference and adds bias. With a neat new design, you can reduce bias with very little effort.

3

11

140

» teej

@teej_m

1 year

Data people – What was the last technical book you found useful?

51

10

139

» teej

@teej_m

2 years

Did you know that queries in Snowflake are translated to a streaming data pipeline? You might have guessed it if you've ever looked at the query profile. All optimization starts at the query profile. It's so much more than an explain plan, take a look:

5

13

137

» teej

@teej_m

1 year

Always cast source timestamps to UTC. That is your canonical value. Then create 3 other columns: UTC date, biz timezone time, biz timezone date. Add suffixes with the grain and tz. No to tz means biz tz. So 4 cols: created_at_utc created_at_date_utc created_at created_at_date

Alex Kravetz

@adkravetz

1 year

@teej_m @HamelHusain @sarahcat21 Can you say more about the timestamp conventions? Running into exactly these sort of issues and want to get ahead.

0

3

6

13

137

» teej

@teej_m

5 months

Your data pipeline when someone upstream drops a column

Crazy Clips

@crazyclipsonly

5 months

Fire destroys aluminum plant in seconds 😳

732

2K

25K

2

16

135

» teej

@teej_m

1 year

If you ever want self serve to work, you need to make a grocery store. You can’t give folks directions to the farm to pick their own produce.

4

11

136

» teej

@teej_m

4 years

Every analytics tool on the planet needs this. ALL OF THEM. This is a huge time saver. Analyst productivity = business productivity. I want tools that make me and my team super productive. I get irritated that no one seems to care about the humans using these tools.

Patrick Collison

@patrickc

4 years

The new @Stripe Dashboard analytics now make it super easy to perform all the usual intertemporal comparisons in a few clicks, automatically surfacing past periods of equal length. Much less fiddling with calendar/date widgets required!

12

34

507

3

9

132

» teej

@teej_m

9 months

@pandas_dev Aww, apology accepted. Look, we’re on the same team, at least we’re not writing JavaScript.

8

3

128

» teej

@teej_m

3 years

SQL 201 - the foundational skills for OLAP Syllabus - CTEs - Self join - Set difference - Basic windows - Unions

4

12

130

» teej

@teej_m

4 years

I’ve been coy before but screw it – I work on Twitch-scale live streaming video. I get to solve problems that have never been solved before. The catch: it’s online sex work The upside: My job helps make the worlds oldest profession safer than ever And I’m hiring a data eng

6

31

128

» teej

@teej_m

3 years

@FrankARinaldi1 @david_perell The premise is bogus. These subjects are all 10th/11th grade classes. Average Harvard applicant in 2020 knows way more than this.

1

0

125

» teej

@teej_m

3 years

You will never reproduce Google Analytics numbers from your own log data. Stop trying to do this. Just don’t do it. Your engineers may convince themselves they can. They cannot.

8

13

121

» teej

@teej_m

2 years

Snowflake announced a new table type - iceberg table. What makes this different than an external table is that it’s read/write, with all the data & metadata still stored in your S3. They demoed a transaction deleting data from both an iceberg and normal table at once.

9

10

121

» teej

@teej_m

2 years

Let’s talk about actual cost issues with Snowflake and how they could be fixed. There’s 3 categories that matter: - Cost allocation - Cost optimization - Clustering & Partitioning Let’s dig in

Tristan Handy

@jthandy

2 years

[1/N] I screwed up over the weekend with one particular link in my most recent Roundup. I want to take a second to own that I did a poor job using my platform and apologize.

6

3

67

8

13

119

» teej

@teej_m

2 years

What’s good in streaming data these days? Spark/Flink/??? Planning out some streaming compute, mostly windowed aggregates. Looking for diverse perspectives and advice. We value: - low maintenance - in Python/JS/SQL - in GCP nice to have Designing for 200k events/min.

34

7

113

» teej

@teej_m

2 years

Data work is glue work. And an important part of that work is the emotional regulation of the company as a whole. Numbers are not enough. The data team can be a voice to say – yes there's uncertainty, but waiting for more data isn't going to change the decision. This is hard.

6

7

116

» teej

@teej_m

9 months

What makes a data analyst great? One skill I think is important that I want to talk about – thinking in distributions aka “just look at the histogram already”. Distributions are so much more informative than just the average.

3

5

111

» teej

@teej_m

2 years

Great question from the @LocalOptimistic slack - how do you stay technical as a manager? Does it matter? My answer - I constantly worry about not being technical enough An unorganized thread for my unorganized system.

3

10

111

» teej

@teej_m

1 year

I still need to do a write up on the $5k Snowflake query my team ran in the fall.

fks

@FredKSchott

1 year

Just accidentally spent $300 on a single BigQuery query AMA

165

132

4K

10

4

112

» teej

@teej_m

1 year

Most important: you need to learn how to do a technical interview. It’s not a quiz, it’s an audition. Silent and technically correct is a failure. Talk, a lot. Ask questions. Walk me through the problem and how you want to solve it. I need to know we can do the work together.

6

3

108

» teej

@teej_m

2 years

Next week at Snowflake Summit, I'm giving a talk on funnel analysis. I just released the code companion to the talk () There's 2 things I want you to show you today –

GitHub - teej/sf-funnels

Contribute to teej/sf-funnels development by creating an account on GitHub.

github.com

4

12

107

» teej

@teej_m

2 years

I don’t trust any data system that doesn’t support joins.

12

2

106

» teej

@teej_m

5 years

I used to think endianess, the way computers put bits of bytes together, was some nonsense word. It turns out it’s based on Gulliver’s Travels. Two factions feud over which end to break the hard-boiled egg, the big-endians vs the small-endians.

3

31

104

» teej

@teej_m

4 years

@EricpNBC12 Link:

3

15

92

» teej

@teej_m

1 year

Ignore the AI hype. I’ve tried it all, it’s not that great. But you know what’s fun as hell? Midjourney. Best $10 subscription of my life. I would cancel Netflix for this. I only keep Discord on my phone to send dumb prompts to Midjourney.

6

2

106

» teej

@teej_m

1 year

@IgorBrigadir all these prompts will be lost in time, like tears in the rain

0

2

104

» teej

@teej_m

5 months

If you want to write documentation that gets used, you need to 🛑STOP🛑 writing like you're Wikipedia It feels good to put your cute little facts into a tidy taxonomy, but that doesn't help you scale knowledge. Instead, organize around: personas, decisions, and actions

6

12

103

» teej

@teej_m

1 year

I've built data transformations in the warehouse for 7+ years. The development environment - if you can even call it that - has always been a mess. SQLMesh is the thing I wish I had on day 1. Look at all this cool stuff that it does out of the box.

Toby Mao

@Captaintobs

1 year

I love how DevOps enables software engineers to safely ship code faster. As a data engineer, I think we deserve something similar: the DataOps revolution. Here’s a 🧵 on why I made SQLMesh, an open source DataOps framework, which launches today

6

25

163

7

102

» teej

@teej_m

2 years

Just spent an hour debugging a SQL issue. It turns out I completely forgot that you can’t concatenate strings with + SQL was a mistake folks, pack it up, we’re done here

10

1

100

» teej

@teej_m

2 years

What is the actual problem that a "data lineage tool" solves? If you say data mesh, you're fired.

65

4

97

» teej

@teej_m

1 year

Notebooks are a workshop. Production systems are the factory. Not everything needs to be put into production. Not everything should be a notebook. You need both. Lean in to the strength of each. When do you hand-mill vs injection mold?

2

3

98

» teej

@teej_m

2 years

Never been happier about a chart in my life

1

6

93

» teej

@teej_m

3 years

@TechEmails Yesterday I liked your tweet (old fashioned, I still use twitter, smile)

1

0

96

» teej

@teej_m

2 years

One day, you too will struggle to find your birth year in the drop down.

5

10

95

» teej

@teej_m

2 years

Snowflake + dbt 1.1.0 performance trick. This has no business working so well, but it does. For large, incremental models with a surrogate unique_key - add one of your surrogate columns to the unique_key config. That's it. This change cut my MERGE times nearly in half:

2

9

97

» teej

@teej_m

1 year

BI (charts with pickers) feels further from solved than ever. Looker is great at 1-2 things but otherwise awful. All the startups are dead. Tableau will never go away. Everything else sucks, nothing is 10x better.

11

4

93

» teej

@teej_m

3 years

In an interview, I won’t ask you what a p-value is. I don’t know what it is. Not well enough to tell a statistician, at least. I can ride a bike safely even though I have no idea how they stay upright (voodoo). Instead we’ll discuss how you make decisions under uncertainty.

5

7

91

» teej

@teej_m

3 years

Learning to drive decisions quickly, a bias to action, is a critical competency for an analyst. Every skill you learn – communication, storytelling, experimentation, metrics design, causal inference – supports this. Here are a few tactics I use to speed up time-to-decision.

Benn Stancil

@bennstancil

3 years

it's friday, and this one might be a real fight: after talking with @borisjabes , i now believe that analysts should measure themselves entirely on how quickly they can convince other people to make a decision.

20

15

95

4

17

92

» teej

@teej_m

2 months

As a representative of Data Twitter, I hereby declare a moratorium on sankey diagrams

8

2

92

» teej

@teej_m

2 years

@benhamner

3

2

89

» teej

@teej_m

3 years

When I deny a request for data, it's often when a PM is frozen in analysis paralysis. Data teams support decisions. That means you need to decide. Earning the authority to deny requests is one of the most important factors to running a world-class data team.

Sean Rose

@seanrose

3 years

One thing I see a lot of product orgs struggle with is the simple ownership of the fact that they’re making choices. There’s never any amount of data, research, backlogs, etc. that is going to decide anything for you.

3

18

195

5

7

88

» teej

@teej_m

23 days

My YC W24 journey in 3 photos – Giving my unconference talk at the start of the batch (IYKYK) My classic YC sign photo, taken before the @paulg @jesslivingston talk Learning from @daltonc on how to stay sane and not die, on the last day of YC

5

2

88

» teej

@teej_m

2 years

I’m late to this party but R + dplyr seems great. I haven’t written R in bit but the tidyverse syntax clicked pretty fast. I like piping. I want something in Python that’s halfway between dbplyr and Malloy. Im cooking up something dumb, per usual.

16

1

88

» teej

@teej_m

2 years

This is why so much tension exists in the DS role. Analytics is considered ops - a service org of dashboard jockeys. The data bitches with no levers on growth. But Data Science is “hard”, so you hire a team of PhDs to solve problems the business doesn’t actually have.

2

4

83

» teej

@teej_m

2 years

Snowflake summit thread list:

2

9

86

» teej

@teej_m

2 years

I got a preview of something new and I'm so excited about it. Lets talk about ✨event analytics✨ I have over a trillion events in my db. Just getting them in is a pain. Analyzing them is harder. I can count click of a button, but WHY do people click at all? Why did they stop?

5

6

86

» teej

@teej_m

2 years

I’ve sat on board meetings, set company OKRs, gone to the leadership off-sites, done annual planning - all the “in the room where it happens” work. None of that gives me the feeling of influence I get from experimentation. Sitting inside the event loop of a company is powerful.

5

12

86

» teej

@teej_m

16 days

Day 1 at the Titan office

5

1

85

» teej

@teej_m

2 years

Yes, I basically built crappy versions of airflow, dbt, mode, fivetran, retool, eppo, & census/hightouch. I am tired of yall stealing my ideas. Just ask next time, I'll give them to you for free.

» teej

@teej_m

2 years

As we enriched our data, demand grew to use it. We built connectors to push data to google sheets, email marketing tools, on-site personalization, Salesforce, and more. For better or worse, SQL was the core of our stack.