Jason Kealey
@softwareengineering.ca
Exited startup founder giving it another whirl; software engineer.
Particularly interested in productizing LLMs to simplify workflows.
Particularly interested in productizing LLMs to simplify workflows.
Pinned
Jason Kealey
@softwareengineering.ca
· Dec 29
Heading into 2025, I must conjure up the same energy as the guy who erroneously left me 8 voicemails asking me to refund the $14 I didn’t bill him.
I’ll be cold calling to validate my go-to-market motions.
This man will serve as my inspiration to make an additional call.
#buildinpublic
I’ll be cold calling to validate my go-to-market motions.
This man will serve as my inspiration to make an additional call.
#buildinpublic
Anyone got open source alternatives to vanna.ai for text to SQL?
The SQL part is pretty good but the plotly charts it recommends are wonky.
The SQL part is pretty good but the plotly charts it recommends are wonky.
Vanna.AI - Personalized AI SQL Agent
vanna.ai
April 22, 2025 at 11:21 PM
Anyone got open source alternatives to vanna.ai for text to SQL?
The SQL part is pretty good but the plotly charts it recommends are wonky.
The SQL part is pretty good but the plotly charts it recommends are wonky.
I’ve also been playing with dbt - and considered sqlmesh instead.
Chose to go deeper with dbt as I feel like sqlmesh’s real value shines when you want to avoid transforming the data both in dev and prod.
… and I’m just prototyping stuff locally to play with different open source BI frontends.
Chose to go deeper with dbt as I feel like sqlmesh’s real value shines when you want to avoid transforming the data both in dev and prod.
… and I’m just prototyping stuff locally to play with different open source BI frontends.
April 16, 2025 at 11:01 AM
I’ve also been playing with dbt - and considered sqlmesh instead.
Chose to go deeper with dbt as I feel like sqlmesh’s real value shines when you want to avoid transforming the data both in dev and prod.
… and I’m just prototyping stuff locally to play with different open source BI frontends.
Chose to go deeper with dbt as I feel like sqlmesh’s real value shines when you want to avoid transforming the data both in dev and prod.
… and I’m just prototyping stuff locally to play with different open source BI frontends.
Last week I did some experimentation with unstructured.io to extract content out of some pdfs.
Also played with github.com/getomni-ai/z... as a more lightweight option.
Overall, vision models do a much better job than classic OCR (ex: tesseract) on tables in docs.
Also played with github.com/getomni-ai/z... as a more lightweight option.
Overall, vision models do a much better job than classic OCR (ex: tesseract) on tables in docs.
April 16, 2025 at 10:57 AM
Last week I did some experimentation with unstructured.io to extract content out of some pdfs.
Also played with github.com/getomni-ai/z... as a more lightweight option.
Overall, vision models do a much better job than classic OCR (ex: tesseract) on tables in docs.
Also played with github.com/getomni-ai/z... as a more lightweight option.
Overall, vision models do a much better job than classic OCR (ex: tesseract) on tables in docs.
Here’s a decent overview of the data pipeline orchestration tools on the market.
dataengineeringcentral.substack.com/p/review-of-...
dataengineeringcentral.substack.com/p/review-of-...
Review of Data Orchestration Landscape
... and honest one ...
dataengineeringcentral.substack.com
April 16, 2025 at 10:53 AM
Here’s a decent overview of the data pipeline orchestration tools on the market.
dataengineeringcentral.substack.com/p/review-of-...
dataengineeringcentral.substack.com/p/review-of-...
I spent some time this week playing with some tools to setup a data pipeline for simple BI and data science.
Played with airbyte, dbt, duckdb and metabase.
Planning on trying lightdash next week.
Trying to avoid hosted data warehouses and use open source for the full chain.
Played with airbyte, dbt, duckdb and metabase.
Planning on trying lightdash next week.
Trying to avoid hosted data warehouses and use open source for the full chain.
April 4, 2025 at 9:20 PM
I spent some time this week playing with some tools to setup a data pipeline for simple BI and data science.
Played with airbyte, dbt, duckdb and metabase.
Planning on trying lightdash next week.
Trying to avoid hosted data warehouses and use open source for the full chain.
Played with airbyte, dbt, duckdb and metabase.
Planning on trying lightdash next week.
Trying to avoid hosted data warehouses and use open source for the full chain.
Interesting compensation model.
Haven't shared @outseta.com's "build your own adventure" compensation model with BlueSky yet.
1) Everyone is paid based on a salary of $210k per year
2) Everyone can work anywhere from 1 to 5 days per week
3) Everyone earns equity on the same terms as the founders
www.outseta.com/posts/outset...
1) Everyone is paid based on a salary of $210k per year
2) Everyone can work anywhere from 1 to 5 days per week
3) Everyone earns equity on the same terms as the founders
www.outseta.com/posts/outset...
Outseta's Choose Your Own Adventure Compensation Model | Outseta
A salary of $210,000 for everyone. Choose to work 1 to 5 days per week. Earn equity on the same terms as our founders. Here's how it all works!
www.outseta.com
March 24, 2025 at 12:17 PM
Interesting compensation model.
Reposted by Jason Kealey
I published some notes on OpenAI's new text-to-speech and speech-to-text models. They're promising, but like other LLM-driven multi-modal models they appear to suffer from the prompt-injection-adjacent problem of mixing instructions and data in the same token stream
simonwillison.net/2025/Mar/20/...
simonwillison.net/2025/Mar/20/...
New audio models from OpenAI, but how much can we rely on them?
OpenAI announced several new audio-related API features today, for both text-to-speech and speech-to-text. They’re very promising new models, but they appear to suffer from the ever-present risk of ac...
simonwillison.net
March 20, 2025 at 8:41 PM
I published some notes on OpenAI's new text-to-speech and speech-to-text models. They're promising, but like other LLM-driven multi-modal models they appear to suffer from the prompt-injection-adjacent problem of mixing instructions and data in the same token stream
simonwillison.net/2025/Mar/20/...
simonwillison.net/2025/Mar/20/...
Love that #econsky is here.
A few years ago, I started listening to podcasts about the economy during my runs and I feel like it massively helped me learn more about how the world works.
A few years ago, I started listening to podcasts about the economy during my runs and I feel like it massively helped me learn more about how the world works.
March 19, 2025 at 1:09 AM
Love that #econsky is here.
A few years ago, I started listening to podcasts about the economy during my runs and I feel like it massively helped me learn more about how the world works.
A few years ago, I started listening to podcasts about the economy during my runs and I feel like it massively helped me learn more about how the world works.
Reposted by Jason Kealey
It’s surprising how much N, and how much separation you need to detect something like a 10% difference in A/B testing results.
And a 10% lift in something (e.g. conversions) is pretty large.
The math:
And a 10% lift in something (e.g. conversions) is pretty large.
The math:
Easy statistics for A/B testing and hamsters
A/B testing tools often lie about whether something is “statistically significant.” Here’s an extremely simple, mathematically sound formula to compute it for yourself.
longform.asmartbear.com
March 16, 2025 at 9:51 AM
It’s surprising how much N, and how much separation you need to detect something like a 10% difference in A/B testing results.
And a 10% lift in something (e.g. conversions) is pretty large.
The math:
And a 10% lift in something (e.g. conversions) is pretty large.
The math:
Has anyone here attempted to use the most recent multimodal models to build a replacement for Azure Document Intelligence or Amazon Textract?
(Something that works on arbitrary forms.)
(Something that works on arbitrary forms.)
March 15, 2025 at 1:18 PM
Has anyone here attempted to use the most recent multimodal models to build a replacement for Azure Document Intelligence or Amazon Textract?
(Something that works on arbitrary forms.)
(Something that works on arbitrary forms.)
This tariff chaos is an opportunity for Canadian entrepreneurs, but also bad actors.
Looked online for new places to buy furnace air filters. Found a Canadian manufacturer. Great price. Sells retail and wholesale.
Problem? I have no idea if they’re legit.
Looked online for new places to buy furnace air filters. Found a Canadian manufacturer. Great price. Sells retail and wholesale.
Problem? I have no idea if they’re legit.
March 9, 2025 at 12:14 AM
This tariff chaos is an opportunity for Canadian entrepreneurs, but also bad actors.
Looked online for new places to buy furnace air filters. Found a Canadian manufacturer. Great price. Sells retail and wholesale.
Problem? I have no idea if they’re legit.
Looked online for new places to buy furnace air filters. Found a Canadian manufacturer. Great price. Sells retail and wholesale.
Problem? I have no idea if they’re legit.
I like seeing this kind of data being shared.
I wonder if that first filter includes applicants from other countries with zero related qualifications for the in-office role.
I wonder if that first filter includes applicants from other countries with zero related qualifications for the in-office role.
Actual numbers on what it took a seed-stage startup to hire their first two founding backend engineers (full-remote)
It's good to know as a candidate: this is the type of competition you might have. And as a founder: this is how much time investment it can take to make a hire
It's good to know as a candidate: this is the type of competition you might have. And as a founder: this is how much time investment it can take to make a hire
March 8, 2025 at 5:24 PM
I like seeing this kind of data being shared.
I wonder if that first filter includes applicants from other countries with zero related qualifications for the in-office role.
I wonder if that first filter includes applicants from other countries with zero related qualifications for the in-office role.
Apparently @profanity.accountant counts swear words instead of beans.
March 7, 2025 at 10:46 PM
Apparently @profanity.accountant counts swear words instead of beans.
This isn’t a rhetorical question, genuinely curious:
Does Amazon make any money when someone buys a dozen CR2032 batteries for five bucks with free next day shipping in a big box?
My best guess is no, but the drop shipper paid monthly fees for the privilege of having items in the warehouse?
Does Amazon make any money when someone buys a dozen CR2032 batteries for five bucks with free next day shipping in a big box?
My best guess is no, but the drop shipper paid monthly fees for the privilege of having items in the warehouse?
March 5, 2025 at 3:47 PM
This isn’t a rhetorical question, genuinely curious:
Does Amazon make any money when someone buys a dozen CR2032 batteries for five bucks with free next day shipping in a big box?
My best guess is no, but the drop shipper paid monthly fees for the privilege of having items in the warehouse?
Does Amazon make any money when someone buys a dozen CR2032 batteries for five bucks with free next day shipping in a big box?
My best guess is no, but the drop shipper paid monthly fees for the privilege of having items in the warehouse?
Reposted by Jason Kealey
Exa Websets — Search, but good
It's an agentic search, like DeepResearch, but instead of producing a 70-page treatise it just gives you a list of pages you never would have found on your own
blog: exa.ai/blog/websets...
try it: websets.exa.ai
It's an agentic search, like DeepResearch, but instead of producing a 70-page treatise it just gives you a list of pages you never would have found on your own
blog: exa.ai/blog/websets...
try it: websets.exa.ai
Exa
The Exa API retrieves the best, realtime data from the web to complement your AI
exa.ai
March 2, 2025 at 12:57 PM
Exa Websets — Search, but good
It's an agentic search, like DeepResearch, but instead of producing a 70-page treatise it just gives you a list of pages you never would have found on your own
blog: exa.ai/blog/websets...
try it: websets.exa.ai
It's an agentic search, like DeepResearch, but instead of producing a 70-page treatise it just gives you a list of pages you never would have found on your own
blog: exa.ai/blog/websets...
try it: websets.exa.ai
It's weird talking to CTOs on both ends of this spectrum.
One end has their inbox flooded with junk applicants.
The other barely has any applicants for legit roles.
Factors: full-remote vs hybrid, tech stack, AI vs no AI, sexy startup vs regular ol' company.
One end has their inbox flooded with junk applicants.
The other barely has any applicants for legit roles.
Factors: full-remote vs hybrid, tech stack, AI vs no AI, sexy startup vs regular ol' company.
Talking to a startup founder of a 5-person, seed-stage company, hiring a sr BE dev:
They tell me LinkedIn job postings don’t work for them
Job postings they put out were overwhelmed with hundreds of applications: most clearly AI-generated (and submitted?). None any good
They tell me LinkedIn job postings don’t work for them
Job postings they put out were overwhelmed with hundreds of applications: most clearly AI-generated (and submitted?). None any good
March 1, 2025 at 3:29 PM
It's weird talking to CTOs on both ends of this spectrum.
One end has their inbox flooded with junk applicants.
The other barely has any applicants for legit roles.
Factors: full-remote vs hybrid, tech stack, AI vs no AI, sexy startup vs regular ol' company.
One end has their inbox flooded with junk applicants.
The other barely has any applicants for legit roles.
Factors: full-remote vs hybrid, tech stack, AI vs no AI, sexy startup vs regular ol' company.
338 is at an Idiocracy-level of ads.
<slowly backs away>
<slowly backs away>
February 26, 2025 at 11:48 AM
338 is at an Idiocracy-level of ads.
<slowly backs away>
<slowly backs away>
This was an interesting read for a few reasons.
1) to Vicky’s point, it appears we’re re-inventing the wheel with an inefficient generalist tool
2) but I’m seeing it as empowering for generalist engineers - as long as they can map complex problems to fundamentals
drive.google.com/file/d/1DsIs...
1) to Vicky’s point, it appears we’re re-inventing the wheel with an inefficient generalist tool
2) but I’m seeing it as empowering for generalist engineers - as long as they can map complex problems to fundamentals
drive.google.com/file/d/1DsIs...
February 26, 2025 at 2:28 AM
This was an interesting read for a few reasons.
1) to Vicky’s point, it appears we’re re-inventing the wheel with an inefficient generalist tool
2) but I’m seeing it as empowering for generalist engineers - as long as they can map complex problems to fundamentals
drive.google.com/file/d/1DsIs...
1) to Vicky’s point, it appears we’re re-inventing the wheel with an inefficient generalist tool
2) but I’m seeing it as empowering for generalist engineers - as long as they can map complex problems to fundamentals
drive.google.com/file/d/1DsIs...
Reposted by Jason Kealey
This is a crazy paper. Fine-tuning a big GPT-4o on a small amount of insecure code or even "bad numbers" (like 666) makes them misaligned in almost everything else. They are more likely to start offering misinformation, spouting anti-human values, and talk about admiring dictators. Why is unclear.
February 25, 2025 at 9:01 PM
This is a crazy paper. Fine-tuning a big GPT-4o on a small amount of insecure code or even "bad numbers" (like 666) makes them misaligned in almost everything else. They are more likely to start offering misinformation, spouting anti-human values, and talk about admiring dictators. Why is unclear.
Within the last six weeks, if stored in optimal conditions, this asset class has appreciated about 60% in the USA.
Follow Charles for more investment insights.
Follow Charles for more investment insights.
Stock markets going down, but I can rest easy knowing I have uncorrelated appreciating assets
February 25, 2025 at 6:00 PM
Within the last six weeks, if stored in optimal conditions, this asset class has appreciated about 60% in the USA.
Follow Charles for more investment insights.
Follow Charles for more investment insights.
Reposted by Jason Kealey
Claude Code
a terminal-based code agent that understands your code base and completes tasks, PRs, etc.
this looks extremely advanced. i may stop using Cursor 🤔
docs.anthropic.com/en/docs/agen...
a terminal-based code agent that understands your code base and completes tasks, PRs, etc.
this looks extremely advanced. i may stop using Cursor 🤔
docs.anthropic.com/en/docs/agen...
Claude Code overview - Anthropic
Learn about Claude Code, an agentic coding tool made by Anthropic. Currently in beta as a research preview.
docs.anthropic.com
February 25, 2025 at 12:25 AM
Claude Code
a terminal-based code agent that understands your code base and completes tasks, PRs, etc.
this looks extremely advanced. i may stop using Cursor 🤔
docs.anthropic.com/en/docs/agen...
a terminal-based code agent that understands your code base and completes tasks, PRs, etc.
this looks extremely advanced. i may stop using Cursor 🤔
docs.anthropic.com/en/docs/agen...
Surprised/happy to discover a French Canadian character in White Lotus s3.
I’m guessing she’ll end up an outright horrible person like all the others in the show. 😂
#quebec
I’m guessing she’ll end up an outright horrible person like all the others in the show. 😂
#quebec
February 24, 2025 at 3:37 AM
Surprised/happy to discover a French Canadian character in White Lotus s3.
I’m guessing she’ll end up an outright horrible person like all the others in the show. 😂
#quebec
I’m guessing she’ll end up an outright horrible person like all the others in the show. 😂
#quebec
I’ve helped a few friends improve the quality of their legacy systems.
Each time, I started by focusing on their data.
From the db, I reversed engineered business rules… and compared that to what they told me the rules were.
…and uncovered several bugs that had gone unnoticed.
Each time, I started by focusing on their data.
From the db, I reversed engineered business rules… and compared that to what they told me the rules were.
…and uncovered several bugs that had gone unnoticed.
February 24, 2025 at 1:51 AM
I’ve helped a few friends improve the quality of their legacy systems.
Each time, I started by focusing on their data.
From the db, I reversed engineered business rules… and compared that to what they told me the rules were.
…and uncovered several bugs that had gone unnoticed.
Each time, I started by focusing on their data.
From the db, I reversed engineered business rules… and compared that to what they told me the rules were.
…and uncovered several bugs that had gone unnoticed.
Always a good laugh when a leadgen company InMails me about ‘challenges’ I must be facing… at the company I sold years ago.
For a business built on outreach, you’d think they’d have better QA.
For a business built on outreach, you’d think they’d have better QA.
February 23, 2025 at 10:14 PM
Always a good laugh when a leadgen company InMails me about ‘challenges’ I must be facing… at the company I sold years ago.
For a business built on outreach, you’d think they’d have better QA.
For a business built on outreach, you’d think they’d have better QA.
Today I learned some people pay a subscription fee for their bed.
And that their bed has a backdoor letting the company ssh into your home network.
Swell!
www.reddit.com/r/technology...
And that their bed has a backdoor letting the company ssh into your home network.
Swell!
www.reddit.com/r/technology...
From the technology community on Reddit: Silicon Valley’s Favorite Mattress, Eight Sleep, had a backdoor to enable company engineers to SSH into any bed
Explore this post and more from the technology community
www.reddit.com
February 23, 2025 at 2:08 AM
Today I learned some people pay a subscription fee for their bed.
And that their bed has a backdoor letting the company ssh into your home network.
Swell!
www.reddit.com/r/technology...
And that their bed has a backdoor letting the company ssh into your home network.
Swell!
www.reddit.com/r/technology...