Need a LLM Kindle feature that is just: “recap what happened in the last few chapters”, for books I pick up after a few weeks away from them.

Still rough around the edges but now have semantic search & summarization for my reading highlights. Working on a write-up next, but TLDR: - Storing OpenAI embeddings in S3. Fine low-cost option for side projects - "Powertools for AWS Lambda" + Serverless framework = 👍 combo

Created an experimental U.S. Web Design System "dev tools" browser extension. It tags each component with a link to the documentation and shows the names of the active breakpoints.

Created a simple browser extension that helps identify which U.S. Web Design System components are in use on a page, and quickly access the official documentation for each component. It also shows what the active responsive breakpoints are. 👉chrome.google.com/webstore/detai…

📝 New post about how I built a semantic search feature for my reading highlights 👇 It turns out you can go surprisingly far with just storing embeddings in a file in S3. sawyerh.medium.com/building-a-scr…

🌅 Hiked to Taft Point and just made it for sunset

Need a phrase for when a document's level of detail crosses a threshold from commendable to infuriating.

I've republished on my personal blog my post about how I built a semantic search feature for my reading highlights. No Medium "paywall" nonsense now. It turns out you can go surprisingly far with just storing embeddings in a file in S3. sawyer.soy/blog/semantic-…

Random experiment with OpenAI text-to-speech API to simulate what the screen reader experience would be with a less robotty voice. OpenAI's voice: drive.google.com/file/d/1iCZyLN… Current VoiceOver voice for comparison: drive.google.com/file/d/1eLhStF…

Refreshed my personal site, adding a bit more detail about the projects I've worked on over the past several years. More exciting: all my blog posts are now centrally available on the Posts page. 🚀 sawyer.soy

Updated my expense email parser to use GPT function calling, replacing the deprecated completion model. Neat to compare the two approaches (function calling is nice!) github.com/sawyerh/expens…

Archetypes are often used to describe a group of users with similar needs, priorities, and general outlooks. They can also be useful when thinking about teams. Here are some of the archetypes I've enjoyed working with when on high-performing teams. sawyer.soy/blog/team-memb…

Experimenting with custom elements ("HTML web components"). Considering creating a package of common dynamic form functionality 🤔 First idea: <conditional-content when="field-name" is="some-value"> Stuff here is conditional </content> sawyer.soy/uploads/condit…

My contribution to today

After reading "The Five Dysfunctions of a Team", I created this visualization of the framework outlined in the book, as a way to synthesize my notes. There's a lot more to it than this, but it can be a useful lens for reflecting and looking at teams through.

👨‍🚀 This week I’m starting a new job, joining Stripe

🦋 You can find me here: bsky.app/profile/sawyer…

Me everytime I get the "Which response do you prefer?" ChatGPT response.