AI Knowledge Sidekick in action + OS repo
Post 3 / 3 of building my personal AI Knowledge sidekick
In the previous two posts, I’ve covered the why1 and the how2.
This time, we get to see what it’s actually like to use this system. Check out the quick overview video at this link.
This has started to change how I consume, connect and stay on top of all the content.
AI analyses all content, with a queryable archive
All content from Readwise is processed and ends up in a queryable archive.
Lots of flexibility for how I can find articles. E.g., keyword, author, tags, content type. This is easily extendible to other filtering criteria.
Each article has a different level of detail. E.g., high-level summaries, tags, and AI-extracted key entities and details. As well as finding similar articles using a semantic similarity search.
Deep dive synthesis
This is the real gold mine. This is where I get the most value.
I stole / borrowed the core ideas of embedding, tagging, clustering, naming clusters & generating insights from each cluster from the Clio system by Anthropic3. This is a worthwhile read if you like making sense of free-text data.
It’s built so I can control which articles go into each synthesis. From a random selection or a specific topic / theme / keyword / author / content type.
Again, it can be extended to other ways of gathering articles for synthesis.
LLM call logs
You can see all the calls made, including the purpose (embed, tag, etc.), the model, input/output tokens, estimated costs, duration, etc.
Gives you essential call log data right in the app.
View and tweak Prompts
This whole app runs on various LLM calls. And sometimes you just want the ability to see the prompts to remember what its doing. Or potentially tweak it for future.
So, I’ve added a page where you do just that right in the app. Making it super convenient.
Semantic Search
And of course, no content database is complete without a semantic search capability. It currently returns the top 50 matching pieces of content, including previous synthesis summaries.
There are future use cases where this can be converted to a chat interface which generates responses from the source content and allows users to ask longer questions.
To the team at Readwise…
Readwise is the one app that I cannot live without. I love the daily highlights emails. The Reader app is the best thing ever; it reminds me so much of Readability, which unfortunately shut down4 in 2016. And I love the chat feature on Reader articles.
But this is the missing piece of the puzzle for me. Something that digs deeper and connects the dots across my entire Reader repo. Steal the idea, and would love to see it in the app.
Over to you to extend the open source repo
Here is the link to the repo or copy and paste this: https://github.com/suhel-nz/Readwise-reader-AI-synthesis
Fork it, make it your own, extend it to your liking. Please fix any bugs you find using your favourite coding agent 😉.
If you’re not a Readwise user, you could probably switch to another source. But you’d need to update the DB schema and ingestion pipeline scripts.
Would love to hear what you do with it.
Final note, Codex is my preferred coding agent, now or atm :)
In the last post, I talked about using Gemini Code Assist in VS Code for all things code. See excerpt below.
AI Agent Software Engineer: Gemini Assist in VS Code was my technical partner in improving, refactoring and fixing bugs. I found it better than GitHub Copilot at considering the whole repo and writing highly performant code. It’s truly amazing what’s possible with LLM coding agents.
But I have completely switched over to OpenAI Codex as the primary AI coding agent. It does a significantly better job than Github Copilot (free), and Gemini Assist (free; yeah, I know I am cheap 😉). But I still use all three when needed to triple-check different approaches to solving problems.
This has been a fun project to wrangle together. And again, it was delayed significantly due to severe illness.
Now it’s onto JobAlchemist full-time (after hours and weekends, obviously). That's been on the back burner for too long, and it's burning a hole in my brain. So, more updates on that are coming soon.
Enjoy!
About me
Hi, I am Suhel.
I am writing to jot down things I’ve learned and noticed on my journey. My posts aren’t intended to be polished, shiny pieces of writing; consider them works in progress. They may be interesting to you; if so, enjoy them and feel free to share if you found them valuable. Feedback is always welcome, please let me know.